What can you run free right now?

Free access

Exact :free model variants list $0 input and output pricing. Accounts with under $10 in lifetime credit purchases receive 20 requests/minute and 50 requests/day; accounts with at least $10 purchased receive 1,000 requests/day.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Child-model disclosure

Every model OpenRouter currently lists as a :free variant. The children below are read from OpenRouter's public model listing each morning, so the roster changes as models enter and leave the free catalog.

• Apodex: Apodex 1.1 Miniapodex/apodex-1.1-mini:free
• Cohere: North Mini Codecohere/north-mini-code:free
• Dots Studio: Dots3-Note Previewdots-studio/dots-3-note-preview:free
• Google: Gemma 4 26B A4Bgoogle/gemma-4-26b-a4b-it:free
• Google: Gemma 4 31Bgoogle/gemma-4-31b-it:free
• LiquidAI: LFM2.5-2.6Bliquid/lfm-2.5-2.6b:free
• NVIDIA: Nemotron 3 Nano Omninvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free
• NVIDIA: Nemotron 3 Supernvidia/nemotron-3-super-120b-a12b:free
• NVIDIA: Nemotron 3 Ultranvidia/nemotron-3-ultra-550b-a55b:free
• NVIDIA: Nemotron 3.5 Content Safetynvidia/nemotron-3.5-content-safety:free
• NVIDIA: Nemotron 3.5 Lightningnvidia/nemotron-3.5-lightning:free
• Poolside: Laguna S 2.1poolside/laguna-s-2.1:free
• Poolside: Laguna XS 2.1poolside/laguna-xs-2.1:free
• Thinking Machines: Inklingthinkingmachines/inkling:free
• Thinking Machines: Inkling Smallthinkingmachines/inkling-small:free

Signup, payment, data

• Signup OpenRouter account and API key required.
• Card not required for the free variants; accounts with under $10 in lifetime purchases get the lower daily limit
• Training Depends on the upstream model provider the request is routed to.
• Terms Varies by routed provider and model; apply OpenRouter privacy/provider filters when data handling matters.
• Clock Standing platform program with a changing model catalog.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “purchase at least 10 credits to raise your daily limit” source
• allowance “| Less than 10 | 20 | 50 |” source
• limits “| Less than 10 | 20 | 50 |” source
• at exhaustion “when you hit one of the platform limits above (free-model requests per minute or per day, or DDoS protection)” source
• deposit unlock “| At least 10 | 20 | 1000 |” source

Free access

Every new account receives 10 trial calls, shared across all free models, with no credit card and no expiry. A one-time top-up of at least $1 permanently switches the account to daily quotas on the free catalog: 100 requests per day, 10 per minute, and one shared pool of 1M tokens per day, resetting daily. Free models bill $0 within these quotas. Some models carry a higher request weight and have lower caps, ranging down to 10 requests per day and 1 per minute.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

60 models from 16 model authors listed at $0 input and $0 output on the provider's free catalog page on 2026-10-09. Nearly all are text models, plus a few image models (gpt-image-2-free, gemini-3.1-flash-image-preview-free). The roster changes over time.

Signup, payment, data

• Signup Account required; no credit card for the initial ten calls. A one-time $1 top-up unlocks the daily quotas.
• Card No credit card is required for signup or the 10 trial calls. The daily free quotas require a one-time top-up of at least $1; the provider does not state which payment methods the top-up accepts.
• Training Not stated by the provider.
• Terms Not stated on the reviewed free-model, quick-start, and console surfaces; check current provider privacy terms before sensitive use.
• Clock Ten trial calls that never expire per new account, then permanent daily quotas after a one-time $1 top-up, reset every day with no published end date.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “Every account starts with 10 trial calls, no credit card, and a $1 top-up switches you to permanent daily quotas.” source
• allowance “All free models share one pool of 1M tokens per day.” source
• limits “coding-glm-5.3-free New Takes text, returns text. 1.05M 100/day · 10/min vs paid →” source

Free access

Explicitly limited free promotion for eligible API organizations that opt in to share API inputs and outputs. Eligible Launch and Grow organizations receive up to 1,000,000 tokens/day in the large-model group and 10,000,000 tokens/day in the small-model group; eligible Build organizations receive 250,000 and 2,500,000 tokens/day respectively. Eligibility is shown on the data sharing settings page, and organizations that do not see the offer there are not eligible. Quotas are shared within each group and reset at 00:00 UTC. A request that would cross the daily quota is billed in full, and overage is billed normally. Only traffic from sharing-enabled projects qualifies. A positive account balance is required. Fine-tuned models, fine-tuning training, evals, and tool use are excluded.

• When it runs out Paid usage starts at normal rates.
• API API compatibility not stated.

Models disclosed

On 2026-10-09, the 1M/250K group listed: gpt-6.1-sol, gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-5.6-sol, gpt-5.5-2026-04-23, gpt-5.4-2026-03-05, gpt-5.2-2025-12-11, gpt-5.1-2025-11-13, gpt-5.1-codex, gpt-5-codex, gpt-5-2025-08-07, gpt-5-chat-latest, gpt-4.1-2025-04-14, gpt-4o-2024-05-13, gpt-4o-2024-08-06, gpt-4o-2024-11-20, o3-2025-04-16, o1-preview-2024-09-12, and o1-2024-12-17. On 2026-10-09, the 10M/2.5M group listed: gpt-5.6-terra, gpt-5.6-luna, gpt-5.4-mini-2026-03-17, gpt-5.4-nano-2026-03-17, gpt-5.1-codex-mini, gpt-5-mini-2025-08-07, gpt-5-nano-2025-08-07, gpt-4.1-mini-2025-04-14, gpt-4.1-nano-2025-04-14, gpt-4o-mini-2024-07-18, o4-mini-2025-04-16, o1-mini-2024-09-12, and codex-mini-latest. The page also keeps gpt-4.5-preview in its list but marks it shut down, so it is not current access.

Signup, payment, data

• Signup required; an OpenAI API organization is required, and only an organization owner can enable sharing
• Card not stated as a card; a positive prepaid balance is required to use the free tokens, and OpenAI's prepaid-billing article puts the minimum initial credit purchase at $5 (payment details are added in that step). The free daily tokens are not billed, so they never draw the balance down.
• Training Provider says it may use your prompts or outputs to train or improve models.
• Terms Yes, explicitly opt-in: shared data is used to identify usage patterns, measure model quality, and inform future evaluation and model training. Sharing is disabled by default and can be turned off at any time.
• Clock No fixed end date is stated. OpenAI promises 30 days' notice before terminating the program; daily quotas reset at 00:00 UTC.
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “You’re entitled to up to 1M (250k for tiers 1-2) tokens per day shared across gpt-5, gpt-5-codex, gpt-5-chat-latest, gpt-4.5-preview (deprecated and shut down as of 7/14/25), gpt-4.1, gpt-4o, o1, o3 and o1-preview models, and up to 10M (…” source
• at exhaustion “If you use more than the combined limit of 1M/250k or 10M/2.5M across these model groups, overage will be billed at normal rates.” source
• payment “Choose your initial credit purchase amount. The minimum purchase is $5, and the default amount is $10.” source

Free access

Free hosted inference while the Inference API is in experimental status; Hetzner says it will notify users in advance by email if that status changes. Rate limits are applied per API key and the API returns HTTP 429 when any of them is exceeded, but as of 2026-09-23 the Inference API docs page no longer publishes the numeric request and token figures. Performance and availability are not guaranteed, requests can queue at peak, and the platform is not for production use.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

On the Inference API docs page (2026-09-17): Qwen/Qwen3.6-35B-A3B-FP8 and Qwen3.8-27B, both 262,144-token context with text and image input. The models endpoint is the definitive list and the selection changes during the experiment; GLM-5.2, DeepSeek-V4-Flash-0731 and Kimi-K2.7-Code were served at launch and dropped when Hetzner paused large models in August 2026.

Signup, payment, data

• Signup required; Hetzner account sign-in to the Hetzner Experiments Platform, then generate an auth token in the Inference tab at experiments.hetzner.com
• Card not required: Hetzner's account-creation guide lists an email address, contact details and a billing address as the requirements, with adding a payment method only an optional later step
• Training Provider says it does not train on your prompts or outputs.
• Terms The Inference API FAQ states: "We do only keep data that is necessary in order to track usage and (in the future) bill usage: request timestamps, token counts etc." It also states: "We do not store the content of request and response and we do not plan on doing so in the future (unless some law requires us to do so)." Hetzner's blog post on the experiment adds, "We do not log any prompt data." No page states anything about using customer data for model training.
• Clock Explicitly an experiment with fewer guarantees than mature products and no committed end date; scope was already reduced once, so treat as volatile.
• Region Hetzner states that it hosts the Experiments platform on its own hardware in its own data centres. The Inference API pages do not state which countries or regions that hardware sits in, and the only EU-based description in Hetzner's material refers to the earlier OpenClaw Experiment rather than the public Inference API. No user-region restriction is stated.

What the provider's page says

• payment “As long as the Inference API remains in experimental status, it is free of charge. Should this status change, we will notify you in advance via email with detailed information.” source
• limits “Timeframe Input Tokens Output Tokens 60s 4M 100k Additionally, we enforce request-level rate-limits as follows: Timeframe Requests 60s 10” source
• at exhaustion “If you exceed any of the rate limits, the API will respond with HTTP Status Code 429.” source

Free access

Poolside offers provider-hosted, OpenAI-compatible chat inference for its Laguna models. The models page says Laguna is free to use for a limited time and shows Laguna S 2.1 and Laguna XS 2.1 under that offer. The Laguna XS 2.1 launch post says free and paid endpoints are both available, with paid pricing of $0.10, $0.20 and $0.05 per 1M input, output and cache-read tokens. No free-endpoint quota or end date is disclosed.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

On 2026-10-09 the models page showed Laguna S 2.1 (poolside/laguna-s-2.1) and Laguna XS 2.1 under the limited-time free offer for the hosted API.

Signup, payment, data

• Signup Poolside API account and key required; provider documentation does not state a paid balance requirement during the promotion.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not fully established on the reviewed promotion/model surfaces; check provider terms before sensitive use.
• Clock Limited-time promotion; no end date or quota is published.
• Region No region restriction stated by the provider.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

Upstage's AI support program gives eligible K-12 schools, universities, university hospitals, nonprofits, and NGOs free access to the provider-hosted Solar Pro LLM and Document Parse for up to one year. Access is application-based and approved per organization; it is free access, not credits or a paid subscription.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Solar Pro 4 (LLM) and Document Parse, as named on the program page on 2026-10-09. Solar Pro 2 and Solar Pro 3 are no longer named in the program, and Upstage lists both for end of service on 2026-10-30 (KST).

Signup, payment, data

• Signup Application and approval required; eligible K-12 schools, universities, university hospitals, nonprofits, and NGOs. Applicants must first sign up for Upstage Console with the same organization email they enter on the application form. After internal review and approval, instructions arrive by email.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not established in the reviewed eligibility/program material; approved organizations should confirm current API data terms before sensitive use.
• Clock Applications are accepted through 2027-03-31 23:00 KST. Free access lasts up to one year and runs from approval until 2027-03-31.
• Region Eligibility is by organization type, not country; no country restriction is stated.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

OpenCode Zen currently prices thirteen models at Free on its pricing table. Twelve are Free for input, output, and cached-read tokens, and Jev 1.13 Free is Free for input and output. The provider describes every free model as available for a limited time and publishes no request quota or end date. Zen's setup flow says to sign in, add billing details, and copy an API key.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

Free rows on the Zen pricing table on 2026-10-09: Big Pickle, Space Bunny Free, LongCat 2.5 Preview Free, Step 5 Preview Free, Exo Free, MiMo-V2.6-Flash Free, MiMo-V2.5 Free, Ling 3.1 Flash Free, Ling 3.0 Flash Fin Free, Nemotron 3 Ultra Free, Nemotron 3.5 Lightning Free, Muse Spark 1.3 Contributor Free, and Jev 1.13 Free. Jev 1.13 Free is a TypeSafe AI System One model that answers typed yes/no, multiple-choice, and rubric questions instead of generating text. Union Alpha Free was no longer listed on 2026-10-09. Muse Spark 1.2 Contributor Free left the free table by 2026-09-17, and Ox Alpha Free and Hy3 Free were removed before 2026-09-01. The Nemotron endpoints are NVIDIA trial endpoints whose use is logged. Other Zen models are pay-as-you-go.

Signup, payment, data

• Signup required: sign in to OpenCode Zen and obtain an API key
• Card required: the documented setup flow says to add billing details before copying the API key
• Training Depends on the upstream model provider the request is routed to.
• Terms Model-specific. Big Pickle, Exo Free, MiMo-V2.6-Flash Free, MiMo-V2.5 Free, Ling 3.1 Flash Free, and Ling 3.0 Flash Fin Free may use collected data to improve their models during the free period. Nemotron 3 Ultra Free and Nemotron 3.5 Lightning Free log use for security and to improve NVIDIA products and services. Muse Spark 1.3 Contributor Free permits prompts and completions to be used to train future Meta models. The providers of Space Bunny Free, LongCat 2.5 Preview Free, and Step 5 Preview Free follow a zero-retention policy and do not use data for model training. Jev prompts and other inputs are not used for training.
• Clock All thirteen are limited-time promotions. No end date is stated for any of them.
• Region All Zen models are hosted in the US and EU. Geographic eligibility restrictions are not stated.

What the provider's page says

• payment “You sign in to OpenCode Zen, add your billing details, and copy your API key.” source

Free access

Accounts without credit get 10 requests per minute and 60 weighted units per day on the free models; accounts holding credit get 20 requests per minute and 120 weighted units per day. The daily counter resets at 00:00 UTC, and longer prompts use more units. The free page says these limits are shared across free models, while the docs page says each free model is counted separately. An additional IP limit, a site-wide free cap of 15 requests per minute, and a limit of 2 concurrent free requests per account also apply. Past the quota, requests continue at the paid rate only when the account holds credit and paid fallback is on (a header can turn that fallback off); otherwise the request gets a 429 until the quota resets. The docs' example rate-limit header (200 requests per minute) is the paid-tier account limit, not the free quota.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

Two current free-quota models on 2026-10-09 (Qwen: Qwen3.7 Flash and DeepSeek: DeepSeek V4 Flash 0731free) plus auto:free routing; the catalog changes over time.

Signup, payment, data

• Signup Free BazaarLink API key required.
• Card Provider says no credit card is required.
• Training Depends on the upstream model provider the request is routed to.
• Terms Provider says it does not store message content by default, processing it in memory and keeping only billing metadata such as token counts, timestamps and model IDs. Every request passes automated content moderation, and the provider says it retains the content of requests judged to violate its policy as compliance evidence. Messages are forwarded to upstream providers and remain subject to their privacy policies.
• Clock Standing gateway program with a changing free catalog.
• Region Taiwan-based service; global availability not established.

What the provider's page says

• payment “Free - No Credit Card Required” source
• allowance “Requests per day 50 / day” source
• limits “Requests per minute 10 / min” source

Free access

Kenari offers free account registration and a Solo tier requiring no card, with 50 requests per day across the current free-model roster (14 models on 2026-10-09) at up to 5 requests per minute. A cumulative top-up of Rp 10,000 raises that to 1,000 requests per day. Its quickstart says users may skip top-up, create a Kenari API key, and call a :free text model through the OpenAI-compatible /v1/chat/completions endpoint without reducing their balance. The free lane is best-effort with no speed or availability guarantee; the daily quota returns HTTP 429 with a Retry-After header when exhausted.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

Fourteen provider-hosted :free text models per Kenari's homepage on 2026-10-09 (it read seventeen on 2026-09-17, eleven on 2026-09-04 and twelve on 2026-08-27); Kenari says the free roster changes over time with the provider catalog.

Signup, payment, data

• Signup Free Kenari account and API key required; users may skip top-up and no card is required for the Solo free lane.
• Card No card is required. Kenari's homepage lists the Solo tier as free with no top up, and its quickstart says no overseas credit card and no wait for approval are needed.
• Training Not stated by the provider.
• Terms Not stated on the reviewed quick-start, billing, and chat documentation; check provider terms before sensitive use.
• Clock Standing Solo tier: 50 requests/day across free models, up to 5 RPM, best-effort with no availability guarantee.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “Tiga langkah dari nol ke request pertama. Tidak perlu kartu kredit luar negeri, tidak perlu menunggu persetujuan.” source
• allowance “Coba gratis tanpa langganan: 50 request model gratis/hari, 5 request/menit.” source
• limits “Coba gratis tanpa langganan: 50 request model gratis/hari, 5 request/menit.” source
• at exhaustion “Habis kuota harian mengembalikan HTTP 429 dengan header Retry-After dan kode alasan free_quota_daily.” source
• deposit unlock “Total top up Rp 10.000 membuka 1000 request/hari” source

Free access

Kilo offers the dynamic kilo-auto/free route with no credits required and several individually identified free models. Free models are available to authenticated and anonymous users. Anonymous use is limited to 200 requests per hour per IP address; upstream providers may impose additional rate limits. Users must configure agentic interactions, autocomplete, and background tasks separately if they want every Kilo inference path to remain free.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Kilo's gateway documentation does not publish a fixed list of free model IDs and directs users to the live model catalog for current free options. Free models are identified by a :free suffix in the model ID, and on 2026-09-22 the authentication documentation gives minimax/minimax-m2.1:free and z-ai/glm-5:free as its examples. kilo-auto/free selects a model dynamically per session from a curated set of available free models, and that mapping updates server-side as free model availability shifts. For an account with no balance, the kilo-auto/small background route resolves to google/gemma-4-26b-a4b-it:free.

Signup, payment, data

• Signup not required for the documented free-model gateway routes; anonymous users are supported
• Card not required for anonymous access; kilo-auto/free explicitly requires no credits
• Training Depends on the upstream model provider the request is routed to.
• Terms Offer-wide warning: Auto Free may route requests to providers that log prompts and outputs and use them to improve services. Model-specific example: NVIDIA free endpoints log use for security and to improve NVIDIA products and services; improvement-session data is described as not linked to identity or a persistent identifier.
• Clock No end date is stated for kilo-auto/free. Individual free-model availability changes over time. NVIDIA endpoints are labeled trial use only, with no stated end date.
• Region Kilo's terms prohibit access from U.S.-embargoed countries and territories and by sanctioned parties. The listed examples include Cuba, Iran, North Korea, Syria, Crimea, and the so-called Donetsk and Luhansk regions of Ukraine.

What the provider's page says

• payment “Free with limited capability. No credits required. The resolved model is selected dynamically per session from a curated set of available free models; the mapping updates server-side as free model availability shifts.” source
• limits “Anonymous users are rate-limited to 200 requests per hour per IP address.” source

Free access

Mancer's live model table shows MythoLite with an input token cost and an output token cost of FREE. Mancer's prices page says that some models are free while most have a per-token credit cost, and it tells users to test on the free models before making a purchase. Free models stay usable even when an account's credit total is negative, so free usage does not depend on buying or holding credits.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

MythoLite is the only model marked FREE for both input and output tokens on Mancer's model table as read on 2026-10-08. It is listed as a LLaMA 2 architecture model at 13B parameters, tagged Creative and Roleplay, and shown as online. Mancer does not state MythoLite's context window or maximum output length. Paid models on the same table include MythoMax, listed at 0.14 credits per input token and 0.24 credits per output token.

Signup, payment, data

• Signup required: Mancer's terms of use direct users to provide accurate information on the platform's registration form and forbid sharing a login, password, or API key, and the pricing page refers to the email used to create the account.
• Card Not stated by the provider.
• Training Trains by default; the provider offers an opt-out.
• Terms By default, Mancer's terms allow Customer Content to provide, maintain, develop, and improve its services, including training its models. Users can opt out of model training and third-party access by following instructions on the platform.
• Clock No time limit or end date is stated for MythoLite's free token price.
• Region Use is subject to applicable local law and trade controls; Mancer prohibits use in or for the benefit of U.S.-embargoed countries or territories and prohibited or restricted parties.

What the provider's page says

• payment “We offer free models, use them to test before making a purchase.” source
• allowance “MythoLite Click for details ONLINE LLaMA 2 | 13B Creative, Roleplay FREE FREE” source

Free access

MegaNova hosts mistralai/Mistral-Small-3.2-24B-Instruct-2506 as a recurring free-quota text model, listed as free on 2026-09-27. Free Tier 1 registration requires no credit card and includes 50 requests per day for this model, within Tier 1 rate limits of 60 requests per minute and 200,000 tokens per minute. Free quota is granted per account and resets daily at 00:00 UTC. MegaNova documents that Tier 2 and above keep free quota access with higher limits. Once the daily free quota runs out, requests normally fail with a quota error unless the account has turned on the optional paid fallback, so this is genuine standing free hosted inference but not unlimited zero-cost usage.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

mistralai/Mistral-Small-3.2-24B-Instruct-2506 is the documented recurring free-quota text model, labeled Free on its model page with a context length of 8,192 tokens on 2026-09-27.

Signup, payment, data

• Signup Free Tier 1 account and API key required; no credit card required. Optional paid fallback must be enabled separately.
• Card MegaNova states that Tier 1 registration is free and requires no credit card.
• Training Not stated by the provider.
• Terms Not stated on the reviewed model, quota, tier, and paid-fallback documentation; check provider terms before sensitive use.
• Clock Standing daily quota: 50 requests per day at Tier 1, resetting at 00:00 UTC. MegaNova documents higher limits for Tier 2 and above.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “Free registration - no credit card required” source
• allowance “Mistral-Small-3.2-24B-Instruct-2506: 50 RPD” source
• limits “RPM: 60 requests per minute TPM: 200,000 tokens per minute” source
• at exhaustion “When a free model's daily quota runs out, requests normally fail with a quota error.” source

Free access

Merge's provider-owned catalog lists nvidia/nemotron-3.5-lightning-30b-a3b as available hosted text inference with one NVIDIA host and Free / Free input and output pricing, as recorded on 2026-09-30. Merge's Free plan (pricing page, 2026-09-04) includes access to all free LLMs, automatic routing with built-in fallback, and starts without a credit card; the earlier $10 first-month credit was removed from the Free tier, while Pro adds $10 in expiring monthly credits and ordinarily charges model cost plus 5%. Merge's Gateway get started page adds that the Free plan covers its list of free models up to a daily request cap, and that adding a card moves the account to Pro and the full catalog. Merge does not publish the size of that daily cap on any of these pages. Merge also warns that its response usage.cost is provider cost rather than necessarily the final invoice, but the exact Nemotron route is explicitly cataloged as free rather than merely lacking pricing metadata.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

nvidia/nemotron-3.5-lightning-30b-a3b through one NVIDIA host at Free / Free input-output pricing.

Signup, payment, data

• Signup Merge account and gateway key required. The model-level free route is separate from Merge plan credits; no paid model usage is required for this route.
• Card not required; the Free plan says 'Start without a credit card'
• Training Not stated by the provider.
• Terms Merge's catalog marks the NVIDIA host for this route 'Zero data retention: No'; neither the catalog nor the gateway pages state whether prompts are used for training, so check both Merge and host policies before sensitive use.
• Clock No published end date for the exact free route; catalog and host availability can change.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “Start without a credit card” source

Free access

The $0 Free plan provides access only to models currently marked free, with standard rate limits and no monthly credit balance.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

Ten catalog entries listed at $0.00/1M on 2026-10-09: inclusionAI Ling 3.0 Flash Fin; inclusionAI Ling 3.0 Flash Sante; inclusionAI Ling 3.1 Flash; Meituan LongCat 2.0; Meituan LongCat 2.5 Preview; Poolside Laguna S 2.1; Poolside Laguna XS 2.1; StepFun Step 3.7 Flash; StepFun Step 5 Preview; Upstage Solar Mini 4. On 2026-10-09 Upstage Solar Pro 4 was no longer listed at $0.00 (now $0.09 in / $0.36 out per 1M), and Tencent Hy3 remained paid at $0.13 in / $0.53 out per 1M.

Signup, payment, data

• Signup Nous Portal account and OAuth login required; provider page does not require a paid plan for free inference.
• Card No card requirement stated for the $0 Free plan.
• Training Trains by default; the provider offers an opt-out.
• Terms Nous Portal's terms of service say Nous Research may use Your Data for research and for developing, training, fine-tuning, and improving AI and machine-learning models, and may make Your Data and Service Data available to third parties for their own use. The terms also say that if you enable Privacy Mode, Nous Research will not store your inference payloads or use them for training, product improvement, or support, except as described in the Privacy Policy. The plan page does not say whether Privacy Mode is available on the Free plan.
• Clock Standing $0 plan; free child-model catalog can change.
• Region Subject to Nous account availability and export restrictions; no narrower public region list recovered.

What the provider's page says

• limits “Free: 50 RPM, 500,000 TPM” source

Free access

OrcaRouter serves selected catalog models at $0 under model ids ending in -free, and the built-in orcarouter/free router covers that free tier behind a single id. Calls to these ids never touch the account wallet, and the allowance is a request rate rather than a balance, so a workspace can keep calling the free tier indefinitely while it stays inside the limits. Free traffic is capped by requests per minute and requests per day. Those ceilings are public, returned for every tier by the GET /api/free-package/public endpoint, and the provider tunes them live. They are tiered by lifetime spend: a workspace that has never topped up gets a deliberately small daily allowance and also a per-request prompt-size cap, which is the one limit the provider does not publish as a number, and crossing the lifetime-spend threshold raises both. Paying is optional and only buys higher limits. Free access additionally requires the workspace owner to have linked a GitHub account that has been registered for some time, unless the workspace has paid enough to be exempt. Model ids without the -free suffix bill at the upstream per-token price, and a free id never falls back to a paid model.

• When it runs out Requests are rejected (429) until the window resets.
• API API compatibility not stated.

Models disclosed

The free lineup is the catalog ids ending in -free plus the orcarouter/free router, which scores each request by difficulty and picks among the workspace's free models. The provider states that the lineup rotates as models join and leave with capacity, so its documentation deliberately lists no ids: the current ids, and the paid model each one shadows, come from the public free-package endpoint. On 2026-10-01 the orcarouter/free model page listed one model in the router, deepseek/deepseek-v4-flash-free, on the light tier.

Signup, payment, data

• Signup An OrcaRouter account and API key are required, and free routes never touch the wallet. Free access is also gated on the workspace owner having linked a GitHub account that has been registered for some time, unless the workspace has paid enough to be exempt. A workspace that meets neither condition has its free requests refused.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not stated on the reviewed routing, billing, and free-model pages; downstream model/provider policies may vary.
• Clock Standing free route with no stated end date. The request-per-minute and request-per-day ceilings are published per tier at the public free-package endpoint and are tuned live, and the lower tier also carries a per-request prompt-size cap that the provider does not publish as a number. The provider describes the free tier as best-effort capacity rather than a guarantee.
• Region No region restriction stated by the provider.

What the provider's page says

• at exhaustion “429 err_free_rate, with retry_after_seconds (also sent as the Retry-After header) A rate window is full. The value is the seconds left in the window that tripped - to the next minute, or to 00:00 UTC. Wait out retry_after_seconds, then…” source

Free access

LLMTR exposes poolside/laguna-xs-2.1 for hosted text inference at $0 input and $0 output. Its documentation says Poolside serves the model free, LLMTR does not deduct from the user's credit balance, and there is no separate token allowance, only LLMTR's standard request-rate limit. The offer may change if Poolside ends free access. Prompts and completions on the free route may be logged and used for product improvement.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

poolside/laguna-xs-2.1 hosted through LLMTR at $0 input/output.

Signup, payment, data

• Signup LLMTR account and API key required; the free route does not deduct from the account credit balance.
• Card Not stated by the provider.
• Training Provider says it may use your prompts or outputs to train or improve models.
• Terms Provider documentation says prompts and completions on the free Poolside route may be logged and used for product improvement.
• Clock No separate token allowance or published end date; availability depends on Poolside continuing the upstream free offer.
• Region No region restriction stated by the provider.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

Free model after a one-time minimum $15 first credit purchase (later top-ups have a $5 minimum), as stated by Together on 2026-10-09; a positive balance is required. Ternary Bonsai 27B lists $0 input and $0 output token prices, so its tokens do not consume the purchased balance under published pricing. Other paid usage can drain the balance, and API access is suspended at zero until credits are added. Rate limits are dynamic and best-effort, not a guaranteed unlimited allowance.

• When it runs out Requests are slowed down.
• API API compatibility not stated.

Models disclosed

Prism-ML/Ternary-Bonsai-27B is listed with zero input and output token charges. Other Together models are not covered by this free listing.

Signup, payment, data

• Signup required: Together account/organization
• Card A payment method and a minimum $15 first credit purchase are required; later purchases have a $5 minimum. A card is not the only documented payment route: Together also documents US bank account (ACH) payments, while a credit or debit card is required for auto-recharge.
• Training Provider says it does not train on your prompts or outputs.
• Terms Together stores prompts and responses by default for product improvements; organization admins can turn storage off to enable zero data retention. Sharing data for model training is opt-in and off by default. Check organization privacy settings before sending sensitive information.
• Clock No end date is stated for the model's zero price. Purchased prepaid credits currently do not expire, but the model catalog and pricing can change.
• Region Serverless endpoints do not offer region selection. Together describes third-party models hosted on its own infrastructure, including North American data centers; no user-country restriction is stated.

What the provider's page says

• payment “Access to the Together platform requires a minimum $5 credit purchase.” source
• at exhaustion “Requests that exceed your limit return a 429 Too Many Requests error.” source

Free access

Regolo's pricing page lists brick-v1-beta at €0.00 input and €0.00 output per 1M tokens under pay-as-you-go, and the models library shows the same model's input and output as Free, so a pay-as-you-go account can call it at a zero token price. Pay-as-you-go carries a €0 monthly fee and bills only for the tokens consumed, which makes this route distinct from Regolo's separate 30-day free trial. Every other model in the library carries a paid per-token rate or is billed per request, per second of audio, per pixel, or per query, so brick-v1-beta is the only zero-cost inference route here. On 2026-10-07 the pricing page also carried a notice that updated rates go live on November 1 and that current pricing rates and plan rules remain active until that date, and the reviewed pages do not state what brick-v1-beta will cost after that change.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

On 2026-10-07 brick-v1-beta was the only model listed at a zero token price. apertus-70b, brick-complexity-pro, gemma4-31b, glm5.2, gpt-oss-120b, gpt-oss-20b, mistral-small-4-119b, qwen3.5-122b, qwen3.5-9b, and qwen3.8-27b all carried paid per-token rates. deepseek-ocr-2, faster-whisper-large-v3, Qwen-Image, Qwen3-Embedding-8B, and Qwen3-Reranker-4B are billed per request, per second of audio, per pixel, or per query rather than per token, so none of them is a free inference route.

Signup, payment, data

• Signup Regolo PAYG account and API key required; PAYG has a €0 monthly fee and brick-v1-beta is listed at €0 input/output.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Regolo presents the service on its pricing page under the heading Zero Data Retention, Fast and Secure AI, and links separate terms, privacy policy, and DPA documents. The reviewed pricing, models, and documentation pages do not state whether prompts sent to brick-v1-beta are used for training, so read Regolo's privacy terms and DPA before sensitive use.
• Clock Regolo lists brick-v1-beta as a beta model and publishes no retirement date for the model itself. On 2026-10-07 the pricing page carried a notice that updated rates go live on November 1, 2026, and that current pricing rates and plan rules remain active until that date, so the €0.00 listing is only stated to hold until then. The reviewed pages do not say what brick-v1-beta will cost after that change, and beta pricing and availability may change.
• Region No region restriction stated by the provider.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

Requesty's pricing and documentation confirm hosted models with no token charges. Enrollment requires a Requesty organization or account, and the Free plan requires no credit card. New organizations receive 200 free-model requests per day and 20 requests per minute across all free models combined; paying organizations receive 1,000 per day and 60 per minute. The daily allowance resets each day, and exceeding a limit returns a rate-limit error without affecting paid models. Requesty says the models are free for now, may change pricing later, and will announce changes in its changelog.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

Requesty's free-models page listed 13 free models on 2026-10-09: NVIDIA Nemotron 3 Ultra 550B, Nemotron 3 Super 120B, Nemotron 3 Nano 30B, Nemotron 3 Nano Omni 30B Reasoning, Nemotron 3.5 Content Safety, and Nemotron 3.5 Lightning 30B; Poolside Laguna M.1 and Laguna XS.2; Google Gemma 4 31B; Mistral Leanstral 1.5; InclusionAI Ling 3.0 Tiny and Ling 3.1 Flash; and Meta Muse Glimmer 30B. The docs page carries a shorter nine-row table. Requesty says the set is free for now and announces changes in its changelog.

Signup, payment, data

• Signup Requesty organization/account required; Free plan requires no credit card.
• Card No credit card is required for the Free plan.
• Training Depends on the upstream model provider the request is routed to.
• Terms Requesty says it never uses your data for training. Retention and training policies of the underlying model providers vary by model, and Requesty shows them on each model detail page; check them before sensitive use.
• Clock Standing daily allowance: 200 free-model requests/day and 20 RPM for new organizations; resets daily and is described as free for now.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “No credit card required” source
• allowance “New organizations 200 20” source
• limits “New organizations 200 20” source
• at exhaustion “Once you hit a limit, requests to free models return a rate limit error until the window resets. Paid models are not affected.” source

Free access

Standing free-model roster on the CN pricing page, which presents itself as syncing prices in real time and showing only available models, usable at no cost within platform rate limits. Rate limits for the free models are fixed values rather than tiered by spending, and free model calls show as a cost of 0 in the account bill. Specific per-model rate limits are published on the model pages. The roster is dynamically maintained and models can be added or removed; the provider states it retains interpretation rights over which models are free.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

Free ('免费') chat models on the CN pricing page on 2026-10-07, several of them behind the page's expand buttons: THUDM/GLM-Z1-9B-0414, THUDM/GLM-4-9B-0414, deepseek-ai/DeepSeek-R1-0528-Qwen3-8B, Qwen/Qwen3.5-4B, Qwen/Qwen3-8B, Qwen/Qwen2.5-7B-Instruct, and XingChenAGI/Xing4.0-29B from ChinaTelecom, which the model centre lists with a release date of 2026-09-17. Also free: tencent/Hunyuan-MT-7B (translation), PaddlePaddle/PaddleOCR-VL-1.5 and deepseek-ai/DeepSeek-OCR (document OCR), BAAI bge embeddings and rerankers, and several ASR and image models. No Baidu ERNIE chat model is free.

Signup, payment, data

• Signup required; account at cloud.siliconflow.cn with login by SMS or email, and an API key created in the console. Identity verification is also required: the platform states that after verifying your identity you can use all free models.
• Card not required for free models; the platform is pay-as-you-go for paid models
• Training Not stated by the provider.
• Terms Not stated by the provider. On the pages read.
• Clock Standing free-model roster with no end date; dynamically maintained, so composition can change.
• Region China platform (siliconflow.cn and api.siliconflow.cn), operated by a Beijing-registered company. The provider also runs a separate international site at siliconflow.com.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

UnoRouter marks part of its catalog free: a model carrying the :free suffix routes only to free providers and never uses an account balance, while the same model name without the suffix is the paid version billed per token. Access requires a UnoRouter account and an API key. UnoRouter's own cap is 1 request per minute, per free model, per user, layered over the upstream provider's own cap, daily token budgets on some free pools that reset at midnight UTC, tokens-per-minute caps triggered by very large prompts, and a per-user concurrency limit on parallel requests. A rate-limited route returns 429 with a Retry-After header, and a model whose free providers are all busy returns 503, which usually clears in minutes. UnoRouter's models, pricing and rate limit pages do not state whether a credit card is required. There is no announced expiry, and availability is best effort: a model leaves the catalog only when every channel is out, and it returns within minutes once a channel recovers.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

Free routes carry a :free suffix in UnoRouter's own catalog, and the set changes as free pools drain or upstream providers change. On 2026-09-28 the newest catalog listings included space-bunny-alpha:free, mimo-v2.6-flash:free, mimo-v2.6-pro:free, axon-1.8-pro:free, axon-1.8-flash:free, axon-1.8-lightning:free, atria-dawn-preview:free, ling-3.0-flash-vl:free, nex-n2.5-mini:free and k2-horizon:free. Not all free models support tool calls or vision, so check the capability badges on the model page.

Signup, payment, data

• Signup A UnoRouter account and an API key are required. UnoRouter's documentation does not state which sign-in methods are available or whether a credit card is collected at signup.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms UnoRouter's models, pricing and rate limit pages do not state how prompts sent to free models are handled or whether they can be used for training. Free requests route across upstream provider groups and can fail over between them, so check UnoRouter's privacy policy and the policy of the provider serving the model before sensitive use.
• Clock Standing best-effort catalog. UnoRouter caps free models at 1 request per minute, per free model, per user, and some free pools also carry daily token budgets that reset at midnight UTC. Upstream provider caps apply on top. UnoRouter publishes no end date.
• Region No region restriction stated by the provider.

What the provider's page says

• limits “Our free-model cap: 1 request per minute, per free model, per user.” source
• at exhaustion “429 A rate limit fired (see the kinds below). Wait for Retry-After seconds, then retry or switch models.” source

Free access

Standing model-specific zero price: input, cached input, cached-input storage, and output are all listed as Free for GLM-4.7-Flash and GLM-4.6V-Flash (and the legacy GLM-4.5-Flash id, which BigModel says now routes to GLM-4.7-Flash). Numeric request/token rate limits are not public on the pricing page and must be checked on the authenticated account rate-limits surface.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

GLM-4.7-Flash is the current free text model and GLM-4.6V-Flash the current free vision-language model on the international pricing page. That page still prints GLM-4.5-Flash at Free on every price dimension, but the China-region BigModel page for GLM-4.5-Flash states the model was retired on 2026-01-30 with requests auto-routed to GLM-4.7-Flash, so treat GLM-4.5-Flash as a legacy id rather than a second free model. Other models' 'Limited-time Free' cached-input storage is not free inference and is not included in this offer. China-region BigModel additionally lists GLM-4-Flash-250414, GLM-4.1V-Thinking-Flash, GLM-4V-Flash, CogView-3-Flash, and CogVideoX-Flash alongside overlapping GLM Flash models.

Signup, payment, data

• Signup required for API use; core services may require an account, login, and API key
• Card Not stated by the provider. The free-model pricing page does not state that a payment method is required
• Training Depends on the upstream model provider the request is routed to.
• Terms Z.AI's Additional Terms for API Services state that End User Content is used only as necessary to provide the API Services, comply with applicable law, enforce policies, and prevent abuse, and that it will not be used to develop or improve Services unless the user explicitly agrees. The same Terms of Use say that for individual users Z.AI reserves the right to process User Content to improve existing Services and develop new products, and that individual users authorize the use and storage of User Content that is not personal data for developing and improving its machine learning and artificial intelligence technologies, so the protection depends on whether the account is treated as a business or developer account rather than an individual one. The China-region BigModel privacy policy separately allows anonymized information to be used for research, statistical analysis, service improvement, and machine learning or model-algorithm training.
• Clock Not stated by the provider. No end date is attached to GLM-4.7-Flash or GLM-4.6V-Flash, the two models the international pricing page lists as Free on every price dimension.
• Region Separate international Z.ai and China-region BigModel endpoints. China API-key functions may require real-name verification. Z.AI's Terms of Use also require users to confirm they are not located in Iran, North Korea, Cuba, Crimea, Donetsk, or Zaporizhzhia.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

ZenMux lists z-ai/glm-4.6v-flash-free and z-ai/glm-4.7-flash-free as active routes with a $0 effective price on both input and output, each marked Free and Rate Limit on 2026-09-28. Access requires a ZenMux account and an API key, created under either the Pay As You Go plan or the paid Builder Plan subscription. The public pages do not state a card requirement or a numeric free-route quota. ZenMux documents that an account must keep a non-negative balance even for free tiers, while some models can require a positive balance. Prior free routes have sunset without becoming permanent, so the lineup is dynamic.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

On 2026-09-23 the free lineup is two routes, both showing Active on their ZenMux model pages: z-ai/glm-4.6v-flash-free, described as the free version of GLM-4.6V, and z-ai/glm-4.7-flash-free, a 30B class model tuned for agentic coding. Each route is served by the Z.ai and BigModel providers at an effective $0 per million tokens for both input and output.

Signup, payment, data

• Signup A ZenMux account and an API key are required. The Quick Start guide has a user sign in with an email address, a GitHub account or a Google account, then choose between Pay As You Go and the Builder Plan subscription, which starts at $20 per month, and then create a key from the matching management page. Account balance must remain non-negative, and the reviewed pages do not prove that a card or a positive funded balance is required for these active free routes.
• Card Not stated by the provider.
• Training Depends on the upstream model provider the request is routed to.
• Terms Varies by routed model/provider; check ZenMux and underlying Z.ai policies before sensitive use.
• Clock No fixed expiry is stated for either route. Both model pages showed Active with no end date on 2026-09-23.
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “$0.021-0.043 $0 / M tokens $0.21-0.43 $0 / M tokens” source

Free access

Free NVIDIA-hosted NIM API endpoints for prototyping, research, development, testing, learning, and evaluation through free NVIDIA Developer Program membership. Production use is excluded. NVIDIA's NIM FAQ says the API catalog is a trial experience, that the rate of requests varies per model and with concurrent users, and that the exact limits are shown in the authenticated account; no public page states a number (the build.nvidia.com FAQ that once said 40 requests per minute now returns 404). Developer Program membership also licenses self-hosted NIM microservices for non-production use on up to 16 GPUs.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Offer-wide hosted API coverage: 160+ leading AI models according to NVIDIA's current NIM FAQ. Self-hosted coverage is broader and is a separate benefit; NVIDIA's current NIM product page describes thousands of models and customizations supported through NIM runtimes.

Signup, payment, data

• Signup required: NVIDIA account and free NVIDIA Developer Program membership
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not stated by the provider.
• Clock No fixed end date stated; access lasts for the duration of NVIDIA Developer Program membership. This is distinct from the separate 90-day NVIDIA AI Enterprise trial.
• Region No region restriction stated by the provider.

What the provider's page says

• No quote-backed numbers yet; the terms above are the record.

Free access

Agnes AI's pricing page lists agnes-2.5-flash and agnes-3.0-flash at $0 per 1M input, output and cached-input tokens. A new user can create an account and API key without subscribing, then call the hosted OpenAI-compatible API. Free/default text access is rate-limited to 10 requests per minute, after the provider cut free and enterprise text rate limits by 50 percent on 2026-09-23. The pricing page says offers can vary by account eligibility or promotion stage and that the account bill is authoritative.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

On 2026-10-04 the text model pricing table lists agnes-2.5-flash and agnes-3.0-flash at $0 input, $0 output and $0 cached input, and the page states that cached input, input tokens and output tokens are currently free for both models. agnes-2.5-pro is billed at its published Pro-tier prices, and agnes-3.0-pro is marked coming soon at those same Pro-tier prices. agnes-2.0-flash is not listed on the pricing page.

Signup, payment, data

• Signup Account and API key required; no subscription is required for the free/default route.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not stated on the reviewed pricing, quick-start, token-plan, and FAQ pages; check provider terms before sensitive use.
• Clock No fixed end date, but the pricing page says eligibility and promotional availability can vary and the account bill is authoritative.
• Region No region restriction stated by the provider.

What the provider's page says

• limits “Text Model default - 30 20” source

Free access

Anonymous AI Endpoints access is available without an API key at 2 requests per minute per IP per model. Authenticated use is a separate billed path with a 400-requests-per-minute per-project per-model limit and requires an OVHcloud Public Cloud project with a payment method. Playground LLM responses are limited to 1,024 output tokens for testing.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

The anonymous limit is documented offer-wide, per model, across the current AI Endpoints catalog. On 2026-10-09 the public catalog listed 20 models, including Llama, Qwen, GPT-OSS, Qwen3Guard moderation, embedding, image-generation, speech-to-text and text-to-speech models. Models announced for retirement remain reachable through the API for a three-month grace period but are hidden from the catalog. On 2026-10-09 OVHcloud listed Mistral-7B-Instruct-v0.3, Mistral-Nemo-Instruct-2407 and Mistral-Small-3.2-24B-Instruct-2506 as being decommissioned. The authenticated catalog also prices some entries at Free outright: on 2026-10-09 these were the beta Qwen3Guard-Gen 0.6B and 8B moderation models, stable-diffusion-xl-base-v10, and four NVIDIA Riva text-to-speech voices. Qwen3.8-27B is paid, at 0.4 EUR per 1M input tokens and 2.7 EUR per 1M output tokens.

Signup, payment, data

• Signup not required for anonymous access; required for authenticated API-key access
• Card not required for anonymous access; a payment method is required for authenticated Public Cloud access, and Discovery-mode projects without one cannot use the service
• Training Provider says it does not train on your prompts or outputs.
• Terms OVHcloud states that user data is never used to train its models and that it keeps only the data strictly necessary for billing. The service is hosted in the European Union, at Gravelines, France.
• Clock No fixed end date for anonymous free access is stated. Individual models have a lifecycle: announced retirements receive a three-month grace period before removal.
• Region No geographic restriction on anonymous use is stated. OVHcloud's worldwide product page lists the Base and Batch APIs with "Worldwide availability (non-EU)"; authenticated project and catalog availability may vary by OVHcloud region.

What the provider's page says

• payment “Anonymous: 2 requests per minute, per IP and per model.” source
• limits “Anonymous: 2 requests per minute, per IP and per model.” source
• at exhaustion “If you exceed this limit, a 429 error code will be returned.” source

Free access

AI Horde is a standing free, community-hosted text and image inference service backed by volunteer compute. It permits anonymous API use with the public key 0000000000 and requires no registration, credit card, or payment. Anonymous jobs receive the lowest queue priority, and the provider states that priority can only be earned by contributing compute and can never be bought or sold. Capacity and model availability depend on the volunteer workers connected at the time. An OpenAI-compatible endpoint at oai.aihorde.net also accepts the anonymous key at the lowest priority, and the provider describes that endpoint as a pilot that might be restricted or expanded in the future.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

Dynamic text-model catalog supplied by volunteer workers; availability depends on currently connected workers.

Signup, payment, data

• Signup No registration required for anonymous access; use the public anonymous API key 0000000000. Registered accounts may earn non-purchasable kudos by contributing compute.
• Card No credit card is required. The provider states that AI Horde can be used for free, even anonymously, with no credit card and no paywall, and that there is no way to buy priority.
• Training Not stated by the provider.
• Terms Generation requests are fulfilled by workers running on computers that volunteers share, so prompts and outputs pass through machines operated by third parties. AI Horde publishes a privacy policy and terms of service, and the same terms and privacy policy apply to the OpenAI-compatible endpoint. The pages reviewed do not state whether prompts or outputs are retained or used to train models.
• Clock Standing community service with no published end date; models, queue time, and capacity vary with volunteer workers.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “Everyone can use AI Horde for free, even anonymously. No credit card, no paywall, and no way to buy priority.” source
• allowance “Everyone can use AI Horde for free, even anonymously. No credit card, no paywall, and no way to buy priority.” source

Free access

15 RPM, 20K TPM, and 20K tokens/day on the default Free tier.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Aion Labs hosted text models listed on its model catalog.

Signup, payment, data

• Signup Aion Labs account required; Free is the default tier.
• Card No top-up required for the Free tier.
• Training Not stated by the provider.
• Terms Not stated by the provider.
• Clock Standing tier according to current provider documentation.
• Region No region restriction stated by the provider.

What the provider's page says

• payment “TIER REQUIREMENT Free Default on signup 1 Any credit top-up” source
• allowance “TIER TOKENS PER MINUTE DAILY TOKEN LIMIT Free 20,000 20,000” source
• limits “Free 15 1 50 2 100 3 200 4 500 5 1000 Token Limits TIER TOKENS PER MINUTE DAILY TOKEN LIMIT Free 20,000 20,000” source

Free access

Standing evaluation/trial API keys are free but limited. Trial keys are capped at 1,000 API calls per month. Chat models are limited to 20 requests per minute each. Other endpoint limits are Audio Transcriptions 5 RPM, Embed 2,000 inputs per minute, Embed Images 5 inputs per minute, EmbedJob 5 RPM, Rerank 10 RPM, Parse 500 RPM, Tokenize 100 RPM, and default endpoints 500 RPM. The documentation states a recurring monthly limit and gives no trial expiration date or one-time credit balance.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Offer-wide FAQ says trial keys can test all Cohere models and APIs. The current model-specific Chat table lists Command A+, Command A Reasoning, Command A Translate, Command A Vision, Command A, Command R+, Command R, Command R7B, and North Mini Code at 20 RPM. Other APIs include Audio Transcriptions, Embed, Embed Images, EmbedJob, Rerank, Tokenize, and endpoints covered by the default limit.

Signup, payment, data

• Signup required; signing up for a Cohere account creates or permits creation of a trial API key
• Card Not stated by the provider.
• Training Trains by default; the provider offers an opt-out.
• Terms Prompts and generations may be used to train Cohere models by default; Cohere says users can opt out in dashboard settings at any time. Cohere's enterprise data commitments and the opt-out FAQ are addressed to paying customers; trial-key use is governed by the general terms and privacy policy.
• Clock Not stated by the provider. The quota is monthly and no fixed trial expiration is stated
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “Trial keys (and prod keys on newer Chat model variants) are limited to 1,000 API calls a month.” source
• limits “Command A 20 req / min 500 req / min” source

Free access

The Nova Terms of Use (last updated December 2, 2025) contain no billing or fee language, and the Amazon Nova Act user guide states that API key authentication through nova.amazon.com/act is free of charge with daily limits, while the separate AWS IAM route bills usage to an AWS account. API keys are issued from the developer surface (Terms 1.5), usage may be subject to rate limits or other restrictions at Amazon's discretion (Terms 1.3), and the Build with Nova schedule states the offering is intended for experimentation and does not support deploying production scale developer applications (S-1.1). No numeric daily limits are published publicly.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Amazon Nova family on the nova.amazon.com developer surface; page metadata names Amazon Nova 2 Lite, Nova Act, and Nova 2 Sonic. On 2026-10-09 the Amazon Nova Act user guide confirmed that Nova Act can be used with an API key from nova.amazon.com/act. The full per-model roster is behind Amazon sign-in and is not publicly confirmable.

Signup, payment, data

• Signup Required: sign in with an Amazon.com account (users must be at least 18), then get an API key from the developer surface, such as nova.amazon.com/act for Nova Act.
• Card Not stated. Neither the Nova Terms of Use nor the Nova Act user guide says whether a payment card is needed; the API key route is described as free of charge and tied to an Amazon.com account.
• Training Provider says it may use your prompts or outputs to train or improve models.
• Terms Interactions are recorded, retained, may be manually reviewed, and are used to provide, develop, and improve Amazon's services including artificial intelligence models (Terms 1.2). Information may be stored on servers outside the user's country (Terms 2.1). No opt-out is stated.
• Clock Standing developer offering. The Nova.amazon.com Terms of Use were last updated December 2, 2025, and Amazon states it may change, suspend, or discontinue Amazon Nova, or any part of it, at any time without notice. No end date is stated.
• Region Amazon Nova is operated in the United States, and Amazon may restrict access to Amazon Nova or certain features from other locations (Terms 2.4). Certain Amazon Nova models or capabilities may not be available in all areas or to all users (Terms 1.1).

What the provider's page says

• payment “Billing: Free of charge with daily limits” source

Free access

Arli AI lists a $0 Free plan with delayed responses that get slower with more requests, a maximum 12K-token context, one request at a time, no image-generation tools, and the exact quota wording "5 times / 2 days trial of all models." The wording does not clearly state whether this means five uses every two days or a one-time two-day trial.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

All text-generation models in Arli AI's current catalog are covered by the free trial allowance. Image-generation models and tools are excluded from the Free plan.

Signup, payment, data

• Signup required: the plan directs users to Sign up free, and Arli prohibits multiple accounts used to evade limits
• Card Not stated by the provider.
• Training Provider says it does not train on your prompts or outputs.
• Terms Arli describes a zero-log approach for content. Its current terms and privacy policy say prompts, inputs, and generated outputs are processed only transiently, immediately discarded, and not stored, retained, or accessed beyond returning the response. This supports no use of prompts or outputs for training. Arli does log usage metadata, including the number of requests sent, the models used, and the parameters of those requests, and retains that metadata for up to 60 days before automatic deletion. It also automatically collects technical data including the user's IP address for security, abuse prevention, and operation of the services.
• Clock The $0 plan has no published plan end date. The model allowance is described ambiguously as "5 times / 2 days trial of all models."
• Region Not stated by the provider. Users must be at least 18 or the age of majority in their jurisdiction and legally authorized to enter the terms

What the provider's page says

• limits “1 request at a time” source
• context cap “Max 12K context tokens” source

Free access

Databricks Free Edition is a no-cost, serverless-only, quota-limited account for personal, academic, learning, experimentation, and other non-commercial or not-for-profit use. It has daily fair-use limits; exceeding quota can shut down compute for the rest of the day and, in extreme cases, the rest of the month. Model serving has a limited number of active endpoints, no GPU serving endpoints, no provisioned throughput, no custom models on GPU or batch inference, and some models are unavailable.

• When it runs out Requests are rejected (429) until the window resets.
• API API compatibility not stated.

Models disclosed

The Free Edition supports experimenting with foundation models and deploying AI systems, but Databricks does not publish a Free-Edition-specific list of included foundation models. The limitation page explicitly says certain models are unavailable.

Signup, payment, data

• Signup required: users choose a signup method and Databricks creates a Free Edition workspace; authentication is limited to email OTP, Google, or Microsoft
• Card not required: Databricks explicitly says no credit card or payment is required
• Training Provider says it may use your prompts or outputs to train or improve models.
• Terms Offer-wide: Databricks explicitly reserves the right to train on Free Edition users' data. This differs from some separately documented Databricks AI assistive features that say prompts and responses are not used to train generative foundation models.
• Clock No expiration. Daily usage limits apply, and extreme overage can suspend compute for the rest of the month. Databricks may delete accounts inactive for a prolonged period.
• Region Available to qualified users who are legally eligible. The Free Edition terms exclude sanctioned persons and persons located or organized in sanctioned countries or territories. A complete supported-country list is not stated.

What the provider's page says

• payment “No credit card or payment required” source
• at exhaustion “If you exceed your quota, your workspace's compute resources will be shut down and unavailable for the rest of the day (and in extreme cases, the rest of the month). While your compute will be unavailable, your data and settings will not…” source

Free access

Standing Gemini Developer API Free tier: standard input and output are marked free of charge for eligible models, subject to project-level RPM, TPM, RPD, and sometimes model-specific IPM or TPD limits. RPD resets at midnight Pacific time. Google no longer publishes one guaranteed numeric quota table: active limits vary by model, project tier, and account status and must be viewed in AI Studio; specified limits are not guaranteed. Google AI Studio usage is free of charge in available regions.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

On 2026-10-07 the Gemini Developer API pricing page marks standard Free tier input and output as free of charge for Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash, Gemini 3.8 Live, Gemini 3.8 Live Extended Thinking, Gemini 3.1 Flash Live Preview, Gemini 3.5 Live Translate, Gemini 3.5 Transcribe Live, Gemini 3.5 Transcribe, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite, Gemini 3.8 Flash TTS, Gemini 3.8 Flash-Lite TTS, Gemini 3.1 Flash TTS Preview, Gemini 3 Flash Preview, Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite, Gemini 2.5 Flash Native Audio, Gemini 2.5 Flash Preview TTS, Gemini Embedding 2, Gemini Robotics ER 2 Preview, Gemini Robotics ER 2 Streaming Preview, and Gemma 4. The pricing page does not list Gemini 2.0 Flash, Gemini 2.0 Flash-Lite, Gemini Embedding, Gemini Robotics ER 1.6 Preview, or Gemini 2.5 Flash-Lite Preview; those names appear only in the Batch enqueued tokens table on the rate limits page, which covers paid tiers. Free access is specific to both the model and the serving mode: Gemini 3.1 Pro Preview, Gemini Omni Flash, the Nano Banana image models, Veo 3.1, and the Lyria music models show Free tier 'Not available,' and the batch, flex, and priority tables are free of charge for some models and not available for others.

Signup, payment, data

• Signup required; a Google account, an active project, and a Gemini API key are required, and the Additional Terms require users to be 18 years of age or older
• Card not required for the Free tier; linking an active billing account is the qualification for paid Tier 1
• Training Provider says it may use your prompts or outputs to train or improve models.
• Terms For Unpaid Services, Google uses submitted content and generated responses to provide, improve, and develop Google products, services, and machine-learning technologies; human reviewers may read and annotate them. In the EEA, Switzerland, and UK, the Paid Services data-use terms apply even to free-of-charge access.
• Clock Google states no end date for the Free tier. Requests per day quotas reset at midnight Pacific time, and Google states that rate limits are subject to change.
• Region Available only in Google's listed supported regions. The Additional Terms state that API Clients made available to users in the European Economic Area, Switzerland, or the United Kingdom may use only Paid Services, although unpaid Gemini API quota and AI Studio access can otherwise be offered there under the stated special data terms.

What the provider's page says

• payment “Free Active project or free trial N/A” source

Free access

Standing organization-level Free Plan with no usage charges and model-specific recurring rate limits. Free Plan table on 2026-10-06: openai/gpt-oss-120b, openai/gpt-oss-20b, and qwen/qwen3.8-27b 30 RPM, 1K RPD, 8K TPM, 200K TPD; openai/gpt-oss-safeguard-20b 3 RPM, 1K RPD, 2K TPM, 200K TPD; meta-llama/llama-prompt-guard-2-22m and meta-llama/llama-prompt-guard-2-86m 30 RPM, 14.4K RPD, 15K TPM, 500K TPD; canopylabs/orpheus-arabic-saudi and canopylabs/orpheus-v1-english 10 RPM, 100 RPD, 1.2K TPM, 3.6K TPD; whisper-large-v3 and whisper-large-v3-turbo 20 RPM, 2K RPD, 7.2K audio seconds per hour, 28.8K audio seconds per day. Rate limits apply at the organization level rather than to individual users, and cached tokens do not count toward them. The page says the exact current limits are on the authenticated account Limits page; a valid payment method is needed only to upgrade to the Developer plan.

• When it runs out Requests are rejected (429) until the window resets.
• API API compatibility not stated.

Models disclosed

Free Plan rate-limit table as of 2026-10-06: openai/gpt-oss-120b, openai/gpt-oss-20b, openai/gpt-oss-safeguard-20b, qwen/qwen3.8-27b, meta-llama/llama-prompt-guard-2-22m, meta-llama/llama-prompt-guard-2-86m, canopylabs/orpheus-arabic-saudi, canopylabs/orpheus-v1-english, whisper-large-v3, and whisper-large-v3-turbo. Removed from the table since 2026-09-01: groq/compound, groq/compound-mini, and qwen/qwen3.6-27b. Removed from the table earlier, since 2026-08-25: llama-3.1-8b-instant, llama-3.3-70b-versatile, llama-4-scout, and llama-guard-4-12b.

Signup, payment, data

• Signup required; a GroqCloud account and API key are required
• Card not required for the Free tier; a valid payment method is required only to upgrade to Developer tier
• Training Not stated by the provider.
• Terms Groq says inference inputs and outputs are not retained by default, with limited retention for persistence features, reliability, or abuse monitoring, and offers zero data retention to all customers. The reviewed surface does not state whether any retained data is used for model training.
• Clock Not stated by the provider. No fixed offer end date is stated
• Region No region restriction stated by the provider.

What the provider's page says

• payment “To upgrade from the Free tier to the Developer tier, you'll need to provide a valid payment method (credit card, US bank account, or SEPA debit account).” source
• allowance “| openai/gpt-oss-120b | 30 | 1K | 8K | 200K | \- | \- |” source
• limits “| openai/gpt-oss-120b | 30 | 1K | 8K | 200K | \- | \- |” source
• at exhaustion “When you exceed rate limits, our API returns a `429 Too Many Requests` HTTP status code.” source

Free access

Awan LLM's Lite plan is labeled Free forever and includes unlimited tokens subject to model context limits, 20 requests per minute, 200 requests per day for small models, 10 requests per day for medium models, and 10 requests per day for large models.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Current small-model catalog: Meta-Llama-3.1-8B-Instruct, Meta-Llama-3-8B-Instruct, Awanllm-Llama-3-8B-Dolfin, and Awanllm-Llama-3-8B-Cumulus. Current large-model catalog: Meta-Llama-3.1-70B-Instruct and Meta-Llama-3-70B-Instruct. The pricing page also assigns 10 requests per day to a medium-model class, but the current models page does not display a medium-model entry.

Signup, payment, data

• Signup required: the provider exposes an account-creation form requiring email and password
• Card Not stated by the provider.
• Training Provider says it does not train on your prompts or outputs.
• Terms Awan says it does not log user prompts or generations, runs its models in-house, and logs only request counts and rates for rate limiting and usage tracking. This supports no prompt or generation use for training.
• Clock No time limit; the plan is labeled Free forever.
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “Unlimited Tokens!” source
• limits “20 req/minute Small Models 200 req/day” source

Free access

Free token access allows 1 request per second, 60 requests per minute, and 250 requests per hour on text generation, with 100,000 tokens per 24 hours counted as input tokens plus output tokens. The provider states that free-token quotas are provided at no charge and may be reduced without notice.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

Current turbo-tier models exposed by the live Models API.

Signup, payment, data

• Signup A free API token from the LLM7 dashboard at dash.llm7.io is required. As of 2026-10-01 the provider documents only two access types, Free token and Pro, so there is no documented anonymous access.
• Card No card requirement stated for anonymous or free-token access.
• Training Not stated by the provider.
• Terms Not stated by the provider.
• Clock Standing access types in current provider documentation.
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “Free token 1,000,000 tokens per 24 hours” source
• limits “Free token 2 40 100” source

Free access

Morph documents a recurring $0 free tier of 200 API requests per month. On 2026-10-02 the pricing page states that each API call to Fast Apply, WarpGrep, Compact, Reflex, Embeddings, or Rerank counts as one request against that 200 per month cap, and that once the cap is exhausted, per-token or plan-credit pricing applies. The same page lists a Free subscription plan at $0 per month carrying 250,000 credits with low rate limits, where one credit is about $0.00001 and credits apply to Fast Apply, WarpGrep, Compact, Reflex, Embeddings, and Rerank. Credits do not roll over: they reset monthly on the subscription date. Morph's hosted chat models are listed with per-token prices rather than inside the free request allowance.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

On 2026-10-02 the free tier covers 200 API requests per month across Fast Apply (morph-v3-fast and morph-v3-large), WarpGrep, Compact, Reflex, Embeddings (morph-embedding-v4), and Rerank (morph-rerank-v4), plus 250,000 monthly credits on the $0 Free plan that apply to those same products. Morph's hosted chat models, including Kimi K3, GLM-5.3, GLM-5.3-Flash, DeepSeek V4.1 Flash, DeepSeek V4 Flash 0731, and Qwen 3.8, are priced per token.

Signup, payment, data

• Signup Morph account and API key required; the documented free tier provides 200 requests/month without a stated card requirement.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not established on the reviewed free-tier and API surfaces; check provider privacy terms before sensitive use.
• Clock Recurring monthly allowance: 200 requests per month, plus 250K credits on the $0 Free plan that reset monthly on the subscription date and do not roll over. No end date published.
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “200 req free every month” source

Free access

Relace offers a hosted API free tier for testing its code-focused models. Signup is required. The API reference states that rate limits are a per-key request budget over a rolling 60 second window, that free accounts start at 15 requests per minute, and that the budget scales with account tier. No free token or request quota, renewal cadence, or expiration date is published.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

On 2026-10-01 the pricing page lists four Relace models with per-token prices: relace-search, relace-compact, relace-apply-3, and relace-rank. The API reference also documents Open Models endpoints for chat completions, messages, and listing models. Provider material does not state which of these models the free tier covers.

Signup, payment, data

• Signup A Relace account and an API key carrying the rlc- prefix are required. Provider material does not state that the free tier itself requires a card or a funded balance, though the API reference lists a 402 response for an account that is out of credits or has no payment method attached.
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not established on the reviewed model/pricing surfaces; check provider privacy and retention terms before sensitive use.
• Clock No expiry or renewal cadence is published.
• Region No region restriction stated by the provider.

What the provider's page says

• limits “Free accounts start at 15 requests per minute, and the budget scales with your account tier up to unlimited for high-volume users.” source

Free access

Terms of Service section 5 and the pricing page both state eligible accounts receive 50,000 free tokens per month, refreshed monthly, intended for developer evaluation and small-scale testing. Free-tier rate limit is 60 requests/min (paid tier 180). Strictly one free-tier account per human or organization; disposable-email domains prohibited.

Reverification due soon; last checked 2026-10-04.

• When it runs out Not stated by the provider.
• API OpenAI-compatible API, per the provider.

Models disclosed

On 2026-10-04 the pricing table marks eight models Eligible for the 50,000 monthly free tokens: deepseek-v4-flash, deepseek-v4-flash-0731, glm-4.7-flash, gpt-oss-120b, gpt-oss-20b, laguna-s-2.1, laguna-xs-2.1, and qwen3.5-9b. The other named models are marked Standard and are paid only: claude-opus-5, claude-sonnet-5, deepseek-v4-pro, glm-5.2, gpt-5.6-terra, gpt-5.6-sol, grok-4.5, kimi-k2.7-code, kimi-k3, and minimax-m2.7. The auto smart router is listed separately with dynamic pricing and a free tier quota that varies by the target model it selects, so an auto request routed to an eligible model can draw on the free allowance. The pricing page labels the catalog dynamic, so the roster can change.

Signup, payment, data

• Signup Required: create an account with your full name, email address, and a password. Signup is completed with email verification and Turnstile CAPTCHA human verification.
• Card not required; top-ups are optional pay-as-you-go via card (Polar.sh) or crypto (OxaPay)
• Training Not stated by the provider.
• Terms User retains ownership of submitted prompts and generated outputs (Terms section 9); no data-training clause present.
• Clock Standing: written into the Terms of Service and pricing page as a recurring monthly allowance, not a promotion.
• Region Not stated by the provider. No region restrictions stated

What the provider's page says

• payment “To use RouteLM, you must create an account by providing a valid email address and secure password.” source
• allowance “Every registered account receives 50,000 free tokens every single month.” source
• limits “Intentionally bypass or attempt to manipulate system rate limits (60 Requests/Min for Free Tier, 180 Requests/Min for Paid Tier).” source

Free access

SEA-LION advertises a free API for prototyping and proof-of-concept development, rate-limited to 10 calls per minute per user. Users create one Trial API Key through the Playground. The provider says the free APIs are not designed for production or high-throughput workloads.

• When it runs out Not stated by the provider.
• API API compatibility not stated.

Models disclosed

The API is offer-wide across the SEA-LION models returned for the user's key. Current documentation explicitly demonstrates SEA-LION v4.5 for instruct and function-calling use, SEA-LION v3.5 reasoning models, aisingapore/SEA-Guard, and SEA-LION-ModernBERT-Embedding-600M. The exact authenticated deployment list must be queried with the generated key and is not enumerated on the public offer page.

Signup, payment, data

• Signup required: sign in to the SEA-LION Playground with a Google account and create an API key
• Card Not stated by the provider.
• Training Not stated by the provider.
• Terms Not stated by the provider. The privacy policy describes collection of account, contact, log, device, browsing, and geographic data, but does not state whether API prompts or outputs are used for model training.
• Clock The key is labeled a Trial API Key, but no expiration date or elapsed-time limit is stated.
• Region Not stated by the provider. The service is operated by AI Singapore/National University of Singapore, but no user-location restriction is stated on the audited offer or API pages

What the provider's page says

• limits “our rate limits is set to 10 requests per minute per user” source

Free access

Standing watsonx.ai Runtime Lite plan with 300,000 foundation-model inference tokens per month, 20 CUH per month for supported machine-learning workloads, 100 pages per month of document text classification and extraction, and a plan-level limit of 2 inference requests per second. Custom foundation models are not available on Lite.

• When it runs out Requests are rejected (429) until the window resets.
• API API compatibility not stated.

Models disclosed

Ready-to-use IBM and third-party foundation models available to the Lite instance, subject to regional and current catalog availability. IBM's supported-model surface describes IBM and third-party models; custom foundation models are not available on Lite. An exact immutable Lite model roster is not stated.

Signup, payment, data

• Signup required: IBM Cloud account and a watsonx.ai Runtime Lite instance
• Card required for a new IBM Cloud account unless the user has a subscription or feature code
• Training Provider says it does not train on your prompts or outputs.
• Terms IBM states that content uploaded to watsonx and foundation-model output are not used to further train or improve IBM foundation models.
• Clock Monthly allowance: IBM states that Lite plan quota limits reset on the first of every month, and that reaching a quota limit suspends the service for the rest of that month. No fixed offer end date is stated.
• Region Service features and model availability depend on the IBM Cloud region selected when provisioning, so the exact eligible models vary by region. IBM also states that a tax identification number is required to create a new personal use account with an address in specific countries or regions, that personal use accounts are not permitted in some locations, and that feature codes are not supported in some countries. A universal user country restriction on the free plan itself is not stated.

What the provider's page says

• payment “For identity verification, a credit card is required when you create a new account.” source
• allowance “Foundation Models: Up to 300,000 tokens per month” source
• at exhaustion “Reaching any quota limit for Lite plan instances suspends the service for that month.” source
• limits “Rate limit per plan ID 2 inference requests per second” source

Free access

Workers Free includes 10,000 Neurons per account per day at no charge. The allocation resets daily at 00:00 UTC. Further operations fail after the limit on Workers Free; using more requires Workers Paid, where usage above the same free allocation is billed at $0.011 per 1,000 Neurons.

• When it runs out Requests are rejected (429) until the window resets.
• API OpenAI-compatible API, per the provider.

Models disclosed

The free Neuron allocation is offer-wide across the Workers AI catalog except models that explicitly require paid billing. The current exclusion list is @cf/moonshotai/kimi-k2.6, @cf/moonshotai/kimi-k2.7-code, @cf/zai-org/glm-5.2, @cf/zai-org/glm-5.3, @cf/zai-org/glm-5.3-flash, @cf/deepseek-ai/deepseek-v4-flash-0731, and @cf/deepseek-ai/deepseek-v4-pro-0813 (as of 2026-09-04); those models can be reached with Workers Paid or prepaid AI Gateway credits. Model-specific Neuron consumption varies by model and input/output type.

Signup, payment, data

• Signup required; a Cloudflare account, Account ID, and API token are required
• Card Not stated by the provider. The provider confirms a Workers Free plan and says paid billing is required only for specified models or overage, but the reviewed pages do not explicitly state the signup card policy
• Training Provider says it does not train on your prompts or outputs.
• Terms Cloudflare says it does not use Customer Content to train Workers AI models or improve Cloudflare or third-party services without explicit consent
• Clock no fixed offer end date stated; the 10,000-Neuron allocation resets daily at 00:00 UTC
• Region No region restriction stated by the provider.

What the provider's page says

• allowance “Our free allocation allows anyone to use a total of 10,000 Neurons per day at no charge.” source
• at exhaustion “All limits reset daily at 00:00 UTC. If you exceed any one of the above limits, further operations will fail with an error.” source