OpenRouter free models: API list, September 2026
OpenRouter's public API priced 21 of 458 models at zero on 28 September 2026. They suit testing, not production: free use is capped at 50 requests a day, or 1,000 after buying at least 10 credits.
On 28 September 2026, OpenRouter’s public models API priced 21 of its 458 listed models at zero for both input and output.1 The longest context window among them is 1,048,576 tokens, shared by four entries: Thinking Machines’ Inkling Small (free) and Inkling (free), and Google’s Lyria 3 Pro Preview and Lyria 3 Clip Preview.1 Free use is capped at 50 requests a day, or 1,000 a day once an account has bought at least 10 credits.2
The free list changes often, so treat it as a dated snapshot, not a standing catalogue.
TL;DR
- 21 models cost nothing per token on 28 September 2026. Eighteen carry the
:freesuffix or are the free router; three are zero-priced listings without the suffix. - Context length runs from 1,048,576 tokens down to 65,536. Eight of the 21 sit at 262,144 tokens.
- NVIDIA supplies five of the 21, the most of any company. Google supplies four.
- The daily cap is the constraint. Fifty requests a day is enough to test a prompt, not to run a product.
- Buying 10 credits lifts the daily cap to 1,000. On a 10-credit top-up, OpenRouter’s purchase fee is the US$0.80 minimum.
- Some free models may keep or train on your prompts. Read each model’s terms before sending anything sensitive.
How this ranking was built
The ranking uses one source and one field. The source is OpenRouter’s public models API at https://openrouter.ai/api/v1/models, read without a key on 28 September 2026.1 Our research standard sets how that capture is kept and checked.
- Fields:
pricing.prompt(input) andpricing.completion(output), which OpenRouter reports in US dollars per token.1 - Inclusion rule: both prices equal zero.1
- Row count: 458 models listed, 21 kept.1
- Ranked field: context length in tokens, as the API reports it. Nothing was merged or removed. Tied entries keep the order the capture gives them.
These are OpenRouter’s listed prices and context lengths. They are reported figures, not measured throughput or tested context.
Which models are free on OpenRouter?
These 21 models were priced at zero for input and output on 28 September 2026, ranked by the API’s context_length figure.1 They sit in the model infrastructure field we cover.
| Rank | Model | Model ID | Context length (tokens) | Listing type |
|---|---|---|---|---|
| 1 | Thinking Machines: Inkling Small (free) | thinkingmachines/inkling-small:free | 1,048,576 | :free variant |
| 2 | Thinking Machines: Inkling (free) | thinkingmachines/inkling:free | 1,048,576 | :free variant |
| 3 | Google: Lyria 3 Pro Preview | google/lyria-3-pro-preview | 1,048,576 | Zero-priced, no suffix |
| 4 | Google: Lyria 3 Clip Preview | google/lyria-3-clip-preview | 1,048,576 | Zero-priced, no suffix |
| 5 | Space Bunny Alpha | stealth/space-bunny-alpha | 1,000,000 | Zero-priced, no suffix |
| 6 | NVIDIA: Nemotron 3.5 Lightning (free) | nvidia/nemotron-3.5-lightning:free | 1,000,000 | :free variant |
| 7 | NVIDIA: Nemotron 3 Ultra (free) | nvidia/nemotron-3-ultra-550b-a55b:free | 1,000,000 | :free variant |
| 8 | Dots Studio: Dots3-Note Preview (free) | dots-studio/dots-3-note-preview:free | 512,000 | :free variant |
| 9 | inclusionAI: Ling 3.0 Flash Sante (free) | inclusionai/ling-3.0-flash-sante:free | 262,144 | :free variant |
| 10 | inclusionAI: Ling 3.0 Flash Fin (free) | inclusionai/ling-3.0-flash-fin:free | 262,144 | :free variant |
| 11 | Qwen: Qwen3.8 27B (free) | qwen/qwen3.8-27b:free | 262,144 | :free variant |
| 12 | Poolside: Laguna S 2.1 (free) | poolside/laguna-s-2.1:free | 262,144 | :free variant |
| 13 | Poolside: Laguna XS 2.1 (free) | poolside/laguna-xs-2.1:free | 262,144 | :free variant |
| 14 | Google: Gemma 4 26B A4B (free) | google/gemma-4-26b-a4b-it:free | 262,144 | :free variant |
| 15 | Google: Gemma 4 31B (free) | google/gemma-4-31b-it:free | 262,144 | :free variant |
| 16 | NVIDIA: Nemotron 3 Super (free) | nvidia/nemotron-3-super-120b-a12b:free | 262,144 | :free variant |
| 17 | Cohere: North Mini Code (free) | cohere/north-mini-code:free | 256,000 | :free variant |
| 18 | NVIDIA: Nemotron 3 Nano Omni (free) | nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free | 256,000 | :free variant |
| 19 | Free Models Router | openrouter/free | 200,000 | Router |
| 20 | NVIDIA: Nemotron 3.5 Content Safety (free) | nvidia/nemotron-3.5-content-safety:free | 128,000 | :free variant |
| 21 | LiquidAI: LFM2.5-2.6B (free) | liquid/lfm-2.5-2.6b:free | 65,536 | :free variant |
Seven entries offer a context window of 1,000,000 tokens or more. The largest cluster is 262,144 tokens, where eight models sit.
By company, NVIDIA leads with five entries: Nemotron 3.5 Lightning, Nemotron 3 Ultra, Nemotron 3 Super, Nemotron 3 Nano Omni and Nemotron 3.5 Content Safety. Google has four, split between two Lyria previews and two Gemma 4 models. Thinking Machines, inclusionAI and Poolside have two each.1
The listing type matters for the limits. OpenRouter’s FAQ describes :free as “a free version of the model with its own rate limits”.2 Lyria 3 Pro Preview, Lyria 3 Clip Preview and Space Bunny Alpha are priced at zero without that suffix, and the FAQ doesn’t say which request caps govern them.
The Free Models Router, openrouter/free, isn’t a model of its own. The FAQ describes it as a way “to automatically select a free model for your requests”.2 Use it when any free model will do. Name a :free model directly when the output has to be repeatable.
What are the limits on OpenRouter’s free models?
OpenRouter caps free-model use at 50 requests a day for accounts that have bought fewer than 10 credits.2 Accounts that have bought at least 10 credits get 1,000 free-model requests a day.2
The cap counts requests, not tokens. A request that fills a 1,000,000-token context window counts once, the same as a one-line prompt. That makes the long-context models in the table the better value per request, if the task needs the room.
OpenRouter’s own FAQ says these models have low rate limits and “are usually not suitable for production use”.2 Fifty requests a day covers testing a prompt, not serving users. The decision changes at 1,000 a day: that covers a personal tool, a demo or a small internal script, but still not a customer-facing product with real traffic.
What unlocks higher limits?
Buying at least 10 credits raises the free-model cap from 50 to 1,000 requests a day.2 The FAQ ties the tier to credits purchased, so the credits don’t have to be spent on paid models to count.
The economics are small but not zero:
- OpenRouter’s credits are denominated in US dollars.2
- It charges a 5.5% fee on credit purchases, with a US$0.80 minimum.2 On a 10-credit purchase, 5.5% is US$0.55, so the US$0.80 minimum applies.
- Refunds for unused credits must be requested within 24 hours of purchase.2
- OpenRouter reserves the right to expire unused credits one year after purchase.2
Beyond 1,000 requests a day, the route is a paid model. OpenRouter lists a price per million tokens for each paid model, usually with different prompt and completion prices.2 What those per-token prices depend on is set out in what controls AI inference cost.
Why do other free-model lists disagree?
Other OpenRouter pages count free models differently because they use different rules, not because the models changed.
- The Free Models collection ranks the top 16 models by tokens processed over the trailing seven days.3 It measures use, not price, and it includes two VoyageAI rerankers listed at $0.02 and $0.05 per million tokens.3
- A search for “free” on the models page returns 18 text models.4 That matches the 18 entries in our table whose name or ID contains “free”. The three zero-priced listings without the word, the two Lyria previews and Space Bunny Alpha, don’t match a text search.
- The models page also shows non-text entries at zero, such as Respan’s Span-01 Lite, listed on 26 September 2026, and LiquidAI’s LFM2.5-Embedding-350M (free).5 They didn’t appear in the models API capture, so they aren’t ranked here.
Context figures can also differ within OpenRouter. Nemotron 3 Super’s description mentions a 1M token context window, while its listing shows 262K.3 Nemotron 3 Nano Omni’s description says it supports up to 300K context, and its listing shows 256K.3 The table uses the API figure, because that is the field the ranking is built on.
What do free models cost besides money?
Some free models are paid for with data. Poolside states that if you use Laguna S 2.1 or Laguna XS 2.1 for free, it may use your inputs and outputs to train its models.3 Space Bunny Alpha is a stealth model from an anonymous provider; its prompts and completions may be retained by that provider, though not used for training.3 LiquidAI’s free embedding model states that requests may be retained and used to train Liquid models.5
The rule for a team is simple. Keep client data, credentials and unreleased code out of free endpoints unless the model’s own terms say it isn’t kept or trained on. Confirm this on each model’s page before the first request.
What the ranking cannot tell you
This list ranks context length, which is capacity, not quality. A 1,048,576-token window says nothing about how well a model reasons across it.
It also fixes one day. The models page shows new listings dated 25 and 26 September 2026, so the free set can change within days.5 The list doesn’t cover speed, uptime or queueing under load. And it doesn’t settle which caps apply to the three zero-priced listings without the :free suffix; OpenRouter’s FAQ doesn’t say.
Frequently asked questions
Does OpenRouter give free models?
Yes. On 28 September 2026, OpenRouter’s models API listed 21 models priced at zero for both input and output.1 They are capped at 50 requests a day, or 1,000 a day after buying at least 10 credits.2
Is OpenRouter itself free to use?
OpenRouter says new users receive a small free allowance to test the service.2 Paid models draw on prepaid credits, and OpenRouter takes a 5.5% fee, with a US$0.80 minimum, when you buy them.2
Which free OpenRouter model has the biggest context window?
Four entries tie at 1,048,576 tokens: Inkling Small (free), Inkling (free), Lyria 3 Pro Preview and Lyria 3 Clip Preview.1 For text work, the two Inkling models are the ones OpenRouter describes as suited to reasoning and coding.3
Are OpenRouter free models any good for production?
OpenRouter says they are usually not suitable for production use.2 The binding constraint is the daily request cap, followed by the data terms some free models attach. Use them to prototype and compare, then move to a paid model when traffic is real.
What are the most affordable OpenRouter models?
The 21 models in the table above cost nothing per token. Beyond them, OpenRouter lists separate prompt and completion prices per million tokens for each paid model.2 Compare paid options on the unit your workload uses most, input or output.
New rankings like this one go out through our email briefing.
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
