# Free AI models you can call by API, October 2026 Free AI models on OpenRouter on 2026-10-13: every free model id with its maker, context and inputs, the caps on free requests, and how to re-check the list. ## Summary On 2026-10-13 the OpenRouter model list held 461 models, and 17 of them showed a token price of zero. Fifteen of those are free AI models you can call by API today: 14 free model variants and one router that picks among them. The other two are music models billed per clip or per song. A "free AI API" search turns up lists that are months old, name models that left, or count trials as free. This list is dated, comes from one read of the live list, and keeps only entries with a zero price for both input and output tokens. You get the full table, what free tiers cap, a checklist for picking a model by job, and the one command that re-reads the list so you never depend on a stale page, this one included. ## Free AI models on OpenRouter, the full list Each row is a model id you can send to OpenRouter's chat API as it stands. Context is the context length in tokens, and the longest answer is the maximum output the list states for the top provider. | Model id | Maker | Context | Longest answer | Inputs | |---|---|---|---|---| | cohere/north-mini-code:free | Cohere | 256,000 | 64,000 | text | | dots-studio/dots-3-note-preview:free | Dots Studio | 512,000 | 460,800 | text, image | | google/gemma-4-26b-a4b-it:free | Google | 262,144 | 32,768 | text, image, video | | google/gemma-4-31b-it:free | Google | 262,144 | 32,768 | text, image, video | | liquid/lfm-2.5-2.6b:free | Liquid AI | 65,536 | 8,192 | text | | nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free | NVIDIA | 256,000 | 65,536 | text, audio, image, video | | nvidia/nemotron-3-super-120b-a12b:free | NVIDIA | 262,144 | 235,929 | text | | nvidia/nemotron-3-ultra-550b-a55b:free | NVIDIA | 1,000,000 | 65,536 | text | | nvidia/nemotron-3.5-content-safety:free | NVIDIA | 128,000 | 8,192 | text, image | | nvidia/nemotron-3.5-lightning:free | NVIDIA | 1,000,000 | 65,536 | text | | poolside/laguna-s-2.1:free | Poolside | 262,144 | 32,768 | text | | poolside/laguna-xs-2.1:free | Poolside | 262,144 | 32,768 | text | | thinkingmachines/inkling:free | Thinking Machines | 1,048,576 | 262,144 | text, image, audio | | thinkingmachines/inkling-small:free | Thinking Machines | 1,048,576 | 262,144 | text, image, audio | | openrouter/free | OpenRouter router | 200,000 | not stated | text, image | Prices read from the OpenRouter model list through looot on 2026-10-13. All fifteen return text. `openrouter/free` is not a model: the list describes it as a router that selects free models at random from those available. Use it when you do not care which model answers, and pin a model id when you do. Two entries had a zero token price and are left out on purpose. Google's Lyria 3 Clip Preview and Lyria 3 Pro Preview are music models, and the descriptions in the list price them per clip and per song, so they are not free to call. ## What free tiers limit Free means no charge per token. It does not mean no limits. These are the ones that bite first. - **Requests per day.** OpenRouter allows 50 requests a day on free model variants until you have bought 10 credits in total, then 1,000 a day, per its [limits page](https://openrouter.ai/docs/api-reference/limits) and its [FAQ](https://openrouter.ai/docs/faq). - **Production use.** OpenRouter's FAQ says free models are usually not suitable for production use. - **Answer length.** Several free variants cap the answer well below their context. The Gemma 4 free variants stop at 32,768 tokens and the smallest Liquid model at 8,192. - **Change without notice.** A free variant is a listing, and listings come and go. The table above is true for 2026-10-13. Re-read the list before you build on a model id. - **Not every vendor has one.** Meta's new Muse Wonder and Google's Gemini 4 Argon have no free variant in the list, and Google's pricing page lists Argon as not available on its free tier. Google also runs a free route of its own: the Gemini API pricing page lists Gemma 4 as free of charge on the free tier ([Google's pricing page](https://ai.google.dev/gemini-api/docs/pricing), read 2026-10-14). The [Gemma 4 API post](https://looot.ai/blog/gemma-4-free-and-cheap-api-options) compares that with the OpenRouter variants. ## Free AI models by job: a checklist Start from the job, then pick from the table. This sorts by what the list states, not by quality, which we did not test. - **You want the longest context.** Inkling and Inkling Small list 1,048,576 tokens, and Nemotron 3 Ultra and Nemotron 3.5 Lightning list 1,000,000. - **You send images.** Dots3-Note Preview, both Gemma 4 variants, Nemotron 3 Nano Omni, Nemotron 3.5 Content Safety, both Inkling models and the free router take image input. - **You send audio.** Nemotron 3 Nano Omni and both Inkling models take audio input. - **You send video.** Both Gemma 4 variants and Nemotron 3 Nano Omni take video input. - **You classify content for safety.** Nemotron 3.5 Content Safety is named for that job. - **You write code.** Cohere's North Mini Code is named for it. We did not test how well it codes. - **You want the smallest model.** LFM2.5 2.6B from Liquid AI is the smallest by parameter count in its name. - **You do not want to choose.** Send `openrouter/free` and accept that the model can change from call to call. ## How to check the free list yourself The list changes, so check it before you rely on it. looot reads the OpenRouter model list through its `openrouter-models` endpoint, and that call is free. With the looot CLI set up, this prints every model id with a zero price for input and output tokens: Then read the description of anything new before you trust it, the way the Lyria entries showed a price per clip or song under a zero token price. We ran this command once on 2026-10-13 for the table above. looot's own chat tool, `openrouter-chat`, accepts 15 listed models, and none of the `:free` variants is among them. To call a free model, use your own OpenRouter key. What looot adds is the rest of the agent: one key for web search, page reading, enrichment and other data calls, with each price shown before the call runs. Browse them in the [public catalog](https://looot.ai/public-catalog), see how billing works on [the pricing page](https://looot.ai/pricing), or try a free looot tool such as the [YouTube transcript tool](https://looot.ai/free-tools/youtube-transcript), which costs you nothing and is limited per hour. For the paid side of this week's launches, the [model price page](https://looot.ai/models) lists prices side by side, and the [Muse Wonder page](https://looot.ai/resources/muse-wonder) covers Meta's new model. What "free" means once an agent, not you, makes the calls is in [free AI API options for agents](https://looot.ai/blog/free-ai-api-for-agents). To try it from a terminal first: Last checked 2026-10-13, from one read of the OpenRouter model list. OpenRouter's limits page and FAQ and Google's pricing page were read on 2026-10-14. ## Questions ### What free AI models can I call by API? On 2026-10-13, fifteen: 14 free variants from Cohere, Dots Studio, Google, Liquid AI, NVIDIA, Poolside and Thinking Machines, plus the `openrouter/free` router. The table in this post lists each id with its context and inputs. ### Is there a free AI API with no limits? No. Free variants on OpenRouter are capped at 50 requests a day until you have bought 10 credits, then 1,000 a day, and many cap the length of an answer. OpenRouter's FAQ calls free models usually not suitable for production use. ### How many free requests a day does OpenRouter allow? 50 requests a day on free model variants if you have bought fewer than 10 credits in total, and 1,000 a day once you have bought at least 10, per OpenRouter's limits page and FAQ. ### Is Gemma 4 free to use by API? Yes. OpenRouter lists free variants of Gemma 4 31B and 26B A4B, and Google's Gemini API free tier lists Gemma 4 as free of charge. Paid Gemma 4 listings also exist for jobs that outgrow the free caps. ### Muse Wonder: free or paid? Paid. Meta's Muse Wonder has no free variant in the OpenRouter list, and neither does Google's Gemini 4 Argon. Their per-token prices are on the [model price page](/models). ### How do I check which AI models are free today? Read the OpenRouter model list and keep the models whose input and output prices are both zero. Through looot that read is free: run `openrouter-models` and filter with the `jq` command in this post. ### What is openrouter/free? It is a router, not a model. The OpenRouter list describes it as selecting free models at random from those available, so two calls can be answered by different models. ### Can I use free AI models in production? OpenRouter's own FAQ says free models are usually not suitable for production use. The caps on requests per day and the chance that a free listing disappears are the two reasons to plan a paid fallback. ## For agents ### Code (bash) ```bash looot run openrouter-models --input '{}' --wait=30 | jq -r '.result.data[] | select(.pricing.prompt=="0" and .pricing.completion=="0") | .id' ``` ### List today's free AI models ```text Use looot to find free AI models I can call by API today. First set up https://looot.ai/skill.md. Run the free endpoint openrouter-models and list every model whose prompt and completion prices are both zero, with its id, context length and input types. Flag any model whose description mentions a price per clip, song or request. Do not run any paid endpoint. ``` ### Code (bash) ```bash npx looot init ```