Free AI API: 17 Models at $0 and Their Limits

According to APITariff, 17 chat models were a free AI API on OpenRouter on 2026-09-29, at $0 input and $0 output, each capped at 50 requests a day, or 1,000 a day after $10 of credit, while paid Mistral Nemo, the lowest output price and lowest daily cost among models with at least 128k context, cost $0.053 a day at 1,000 requests. We fetched the OpenRouter model list on 2026-09-29 at 07:46 UTC and counted 17 chat models from 11 providers, with context windows from 65,536 to 1,048,576 tokens. The count is a snapshot: 9 more rows priced $0 in earlier fetches, the oldest from 2026-09-02, did not appear that day. The cap is 20 requests per minute on top of the daily limit, at both credit levels (OpenRouter docs, read 2026-09-29). This page covers OpenRouter. Groq’s free plan and Google’s free-tier rate limits were not read, and the data handling of the 17 models is unconfirmed. $10 of credit covers 188 days at the Mistral Nemo rate ($0.019 input and $0.030 output per 1M tokens, fetched 2026-09-29).

Last updated 2026-09-29 · 17 models · scope: OpenRouter, Google free tier noted, Groq not covered · source: APITariff fetch of OpenRouter, OpenRouter and Google docs read 2026-09-29.

Which free AI API models are available right now?

17 chat models, sorted by context window, largest first. A $0 price does not rank models against each other, so context length is the only ordering rule here. 5 of the 17 have a 1M-token window, 15 have at least 256k, and the median is 262,144.

#	Model	Provider	Context (tokens)	Paid twin, in / out $ per 1M
1	thinkingmachines/inkling-small	Thinkingmachines	1,048,576	0.45 / 1.20
2	thinkingmachines/inkling	Thinkingmachines	1,048,576	1.00 / 4.05
3	nvidia/nemotron-3-ultra-550b-a55b	NVIDIA	1,000,000	0.60 / 2.40
4	nvidia/nemotron-3.5-lightning	NVIDIA	1,000,000	0.06 / 0.16
5	stealth/space-bunny-alpha	Stealth	1,000,000	none
6	dots-studio/dots-3-note-preview	Dots Studio	512,000	none
7	google/gemma-4-26b-a4b-it	Google	262,144	0.09 / 0.30
8	google/gemma-4-31b-it	Google	262,144	0.09 / 0.34
9	inclusionai/ling-3.0-flash-sante	inclusionAI	262,144	none
10	nvidia/nemotron-3-super-120b-a12b	NVIDIA	262,144	0.08 / 0.45
11	poolside/laguna-s-2.1	Poolside	262,144	0.09 / 0.18
12	poolside/laguna-xs-2.1	Poolside	262,144	0.06 / 0.12
13	qwen/qwen3.8-27b	Qwen	262,144	0.0749 / 4.40
14	cohere/north-mini-code	Cohere	256,000	none
15	nvidia/nemotron-3-nano-omni-30b-a3b-reasoning	NVIDIA	256,000	none
16	openrouter/free	OpenRouter	200,000	none
17	liquid/lfm-2.5-2.6b	Liquid	65,536	none

17 of 29 rows shown. 9 rows priced $0 in earlier fetches were absent from the 2026-09-29 fetch and are not counted. 2 Lyria rows are music models and 1 NVIDIA row, Nemotron 3.5 Content Safety, is a safety classifier, so none of the three is a chat model. The paid twin is the same model without the :free suffix; 10 of the 17 have one. The provider of row 5 is listed as Stealth, so we cannot say who runs it. Source: APITariff fetch of openrouter.ai/models, 2026-09-29 07:46 UTC.

Which free AI API has the largest context window?

Two: thinkingmachines/inkling and thinkingmachines/inkling-small, both at 1,048,576 tokens on 2026-09-29. Three more sit at 1,000,000: two NVIDIA models and one Stealth model. The smallest is liquid/lfm-2.5-2.6b at 65,536 tokens, one sixteenth of the largest.

How long does a free AI API stay free?

9 rows that were priced $0 in an earlier fetch were absent on 2026-09-29. The oldest were last seen on 2026-09-02 (minimax/minimax-m3 and minimax/minimax-m2.7) and the newest on 2026-09-28 (inclusionai/ling-3.0-flash-fin). Absent means not in the fetch. Our data does not show whether each was removed, renamed or repriced. 10 of the 17 current models have a paid twin at the price in the table, so a model that leaves the free list may still be available at that price.

How many requests does a free AI API allow?

On OpenRouter, 50 a day by default and 1,000 a day after $10 of purchased credit, with a limit of 20 a minute for all :free models (OpenRouter docs, read 2026-09-29). At 20 a minute the ceiling would be 28,800 requests a day, so the daily cap is the limit that binds, and the $10 threshold multiplies it by 20. Outside OpenRouter, Google’s Gemini API pricing page, last updated 2026-09-24, lists a free tier for input and output tokens on Gemini 2.5 Pro, 2.5 Flash and 2.5 Flash-Lite, and on several Gemini 3.x Flash models. Its rate limits are not covered here, because we did not read them. Groq’s rate-limit page did not show its free-plan table to our fetch, so Groq is not covered either.

Is a free AI API safe for private data?

Treat free requests as public. Google’s pricing page says free-tier content is “used to improve our products”, and paid-tier content is not. OpenRouter’s privacy page, read 2026-09-29, has one account setting for paid models and another for free models. The setting controls whether requests may route to providers that train on your data. The page does not say what each free provider does with prompts. We could not confirm the data handling of any of the 17 models above, so the table has no privacy column. That is a gap in the table, not a finding about the models.

What does a free AI API cost once the free limit runs out?

1,000 requests a day is the free cap for an account with $10 of credit. The same volume on a paid model, at 2,000 input and 500 output tokens a request, is 2M input and 0.5M output tokens a day. The formula is (2 x input price + 0.5 x output price) per day, with prices per 1M tokens.

Model	In / out $ per 1M	Per day	30 days
mistralai/mistral-nemo (131,072 context, not free)	0.019 / 0.030	$0.053	$1.59
poolside/laguna-xs-2.1	0.06 / 0.120	$0.180	$5.40
nvidia/nemotron-3.5-lightning	0.06 / 0.160	$0.200	$6.00
poolside/laguna-s-2.1	0.09 / 0.180	$0.270	$8.10
google/gemma-4-26b-a4b-it	0.09 / 0.300	$0.330	$9.90
google/gemma-4-31b-it	0.09 / 0.340	$0.350	$10.50
nvidia/nemotron-3-super-120b-a12b	0.08 / 0.450	$0.385	$11.55
thinkingmachines/inkling-small	0.45 / 1.200	$1.500	$45.00
qwen/qwen3.8-27b	0.0749 / 4.400	$2.350	$70.49
nvidia/nemotron-3-ultra-550b-a55b	0.6 / 2.400	$2.400	$72.00
thinkingmachines/inkling	1 / 4.050	$4.025	$120.75

11 models shown: all 10 paid twins from the first table, sorted by daily cost, and the first row. The first row is the lowest output price among OpenRouter text models with at least 128,000 context, fetched 2026-09-29, excluding :batch rows; it is also the lowest daily cost at this workload. Source: APITariff fetch of openrouter.ai/models, 2026-09-29 07:46 UTC.

A $10 credit covers 188 days at the Mistral Nemo rate and 28 days at the gemma-4-31b-it rate.

If you want a paid reseller instead, kie.ai has 40 text rows on the kie.ai price page. Its lowest output price is gpt-6-luna at $0.03 input and $0.15 output per 1M tokens, fetched 2026-09-29. That is $0.135 a day at the volume above, about 2.5 times the $0.053 for Mistral Nemo. For video models, the Seedance API pricing comparison sets kie.ai’s rates against Replicate and fal.ai.

Free requests go quickly for an agent. If each step is one request, 50 requests a day are 6 runs of 8 steps (AI agent cost). Put your own token counts through the token calculator to see the daily cost at your volume. The paid rates for the Qwen family behind row 13 of the table, on Alibaba Cloud and OpenRouter, are in the Qwen API pricing guide, read 2026-10-03.

Cite as: According to APITariff (apitariff.com/guides/free-ai-api/, read 2026-09-29).

Drafted with AI from APITariff’s own price data, checked and edited by a person. The chart is rendered from the tables above, not generated.

Change log

• 2026-09-29First edition. 17 free chat models counted from the 2026-09-29 fetch; OpenRouter limits and Google free tier read the same day.