Free LLM APIs in November 2026: which AI APIs are still free

As of November 2026, three LLM APIs give you ongoing free access to capable models: OpenRouter’s free models, Groq’s free plan and Google’s Gemini API free tier. Each has daily limits, and each handles your data differently, so “free” is not the whole story. Several services that used to top “free LLM API” lists have changed: GitHub Models was retired in July 2026, Hugging Face no longer gives free accounts monthly credits, Cerebras offers a short trial that needs a card, and Mistral’s newest docs say its free mode has no API access.

This guide checks each one against its own documentation as of November 2026, says what you get and what you give up, and shows how to use the free ones from an iPhone, iPad or Mac.

Free LLM APIs at a glance

Provider	What’s free	Free limits	Your data on the free tier
OpenRouter	Models whose ID ends in :free (14 text models when we checked)	20 requests a minute, 50 a day; 1,000 a day once you’ve bought at least 10 credits	OpenRouter doesn’t store prompts unless you opt in; free endpoints have their own provider policies
Groq	gpt-oss-120b, gpt-oss-20b, Qwen 3.8 27B (plus Whisper for speech)	Per model: 30 requests a minute, 1,000 a day, 8K tokens a minute, 200K a day	Not retained by default; may be logged up to 30 days for troubleshooting or abuse
Google Gemini API	Several Gemini Flash models, Gemini 4 Hydrogen and Gemma 4	Per project; shown in Google AI Studio	Used to improve Google’s products; human reviewers may read it
Mistral	Free mode covers Le Chat and testing; API keys need pay-as-you-go	Not published	Free mode data may be used for training unless you opt out
Cerebras	$5 trial credits for 30 days, after adding a payment method	5 requests a minute during the trial	Not retained, per its privacy policy
Hugging Face	No free credits for free accounts; PRO includes $2 a month	Pay as you go	Not stored for training, per its docs
GitHub Models	Retired on 30 July 2026	n/a	n/a

Sources: OpenRouter limits, Groq rate limits, Gemini API pricing, Mistral usage limits, Cerebras rate limits, Hugging Face pricing, GitHub changelog, all as of November 2026. Free tiers change often; check the provider’s page before you build on one.

OpenRouter: free models from many companies, one key

OpenRouter is a single API in front of hundreds of models from many providers. Some of them are free: their IDs end in :free and their price is $0 for input and output. When we checked in November 2026 there were 14 free text models, including Google’s Gemma 4 31B and Gemma 4 26B A4B, NVIDIA’s Nemotron 3 models and Liquid AI’s LFM3 1.7B, plus a free router that picks one for you. The list changes from week to week (OpenRouter models).

• Limits: 20 requests a minute. 50 requests a day if you’ve bought fewer than 10 credits in total, 1,000 a day once you have (OpenRouter).
• Data: OpenRouter says it doesn’t store your prompts or responses unless you opt in, and that it doesn’t train on your data (OpenRouter). The model providers behind free endpoints have their own policies, and your account has separate privacy settings for free models (OpenRouter).
• In Modelbase: built in. Sign in with OpenRouter or paste a key, then filter the model list by Free.

Our OpenRouter free models guide covers it in detail.

Groq: fast inference, a free plan with daily limits

Groq runs open models on its own chips and has a free plan. In November 2026 the free plan covers three chat models, each with its own limits (Groq):

Model ID	Requests/min	Requests/day	Tokens/min	Tokens/day
openai/gpt-oss-120b	30	1,000	8,000	200,000
openai/gpt-oss-20b	30	1,000	8,000	200,000
qwen/qwen3.8-27b	30	1,000	8,000	200,000

Older favorites are gone from the free plan: Llama 3.1 8B Instant and Llama 3.3 70B were moved to enterprise customers in August 2026, and several others were shut down (Groq deprecations). Groq asks for a payment method only to upgrade to its paid Developer plan (Groq). By default it doesn’t retain inference data, with an exception for troubleshooting and abuse investigations of up to 30 days (Groq).

Groq’s API is OpenAI-compatible, so it works in Modelbase as a custom server. See our Groq free tier guide for the steps.

Google Gemini API: free tier, with your data as the price

Google’s Gemini API has a free tier: “free of charge” input and output on models such as Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 4 Hydrogen, and on Gemma 4 (Google). Limits apply per Google Cloud project and are shown in Google AI Studio rather than listed in the docs (Google).

Two things to know before you use it:

• Free-tier data trains Google’s products. Google’s terms say it uses what you submit to the unpaid services to improve its products, that human reviewers may read it, and that you shouldn’t submit sensitive, confidential or personal information (Gemini API terms).
• It’s meant for developers. The terms describe AI Studio and the Gemini API as being for developers building “for professional or business purposes, not for consumer use”. If you just want to chat with Gemini, Google’s own Gemini app is the consumer product.

The Gemini API has an OpenAI-compatible endpoint (in beta), which is how Modelbase connects to it. Details are in our Gemini API free tier guide.

What changed: Mistral, Cerebras, Hugging Face and GitHub Models

• Mistral. Mistral’s admin docs say its free mode gives a monthly budget to try its products but no API access; creating API keys needs billing and pay-as-you-go (Mistral). On the free mode, Mistral may use your inputs and outputs to train its models unless you turn that off (Mistral).
• Cerebras. New accounts get $5 of credits for 30 days after adding a verified payment method, and the FAQ says there is no permanent free tier (Cerebras).
• Hugging Face. Inference Providers give free accounts no monthly credits; PRO accounts get $2 a month (Hugging Face).
• GitHub Models was fully retired on 30 July 2026 (GitHub).

All three paid ones still have OpenAI-compatible APIs, so you can connect them to Modelbase the same way if you pay for them.

How to use a free LLM API on iPhone, iPad and Mac

Modelbase is a free app that runs AI models on your iPhone, iPad or Mac, and can also connect to models over an API: OpenRouter with your own key, or any server with an OpenAI-compatible API. Those models appear in the same model picker as the ones on your device.

OpenRouter

1. Tap the model’s name at the top of a chat, then Manage models.
2. Tap + and choose OpenRouter.
3. Tap Sign in with OpenRouter, or paste a key under Or Use a Key (Get a key opens OpenRouter’s key page).
4. Back in Manage models, open OpenRouter, turn on the Free filter and add the models you want.

Groq, Gemini and other OpenAI-compatible APIs

1. Create an API key on the provider’s site.
2. In Modelbase, open Manage models, tap + and choose Custom Server.
3. Enter a Name, the provider’s full API Address and your API Key:

Provider	Address
Groq	https://api.groq.com/openai/v1
Google Gemini API	https://generativelanguage.googleapis.com/v1beta/openai
Mistral (paid)	https://api.mistral.ai/v1
Cerebras (trial or paid)	https://api.cerebras.ai/v1
Hugging Face (credits)	https://router.huggingface.co/v1

4. Tap Test Connection. When it says Connected, the provider’s models appear under Models on This Server; tap the ones you want. (You can also type a model’s ID under Add Model by ID.)
5. Tap Save. The models show up in the model picker under Your servers.

Type the whole address, including the path: Modelbase adds /v1 only when an address has no path at all. Your key is kept in the device’s Keychain, and with iCloud sync on, your providers and keys sync to your other devices through your own iCloud.

For each model you add from a server, you can say whether it accepts images, whether it can use tools (then it can search the web and open pages by itself) and its context length, because such servers don’t report those.

What “free” costs you

• Limits. 50 to 1,000 requests a day is plenty for a person chatting, but a busy app or a long agent session can use it up in minutes. When you hit a limit, the API refuses requests until it resets.
• Your data. A cloud model receives your messages and the images you attach. That’s true of every API here; the difference is what the provider may do with them. Google’s free tier, Mistral’s free mode and some free OpenRouter endpoints may use your prompts to improve their models.
• Availability. Free models come and go. One of OpenRouter’s free models was due to expire on the day we checked, and Groq retired several popular ones this year.

The free AI with no limits: the model on your device

There’s one kind of LLM that is free with no rate limits, no key and no data leaving the device: a model that runs on your own hardware. Modelbase comes with LFM 2.5 (1.2B) inside and has 53 open models to download, from Qwen, Google, Meta, Microsoft, Mistral AI, Liquid AI and others. They’re smaller than cloud models, but they work offline and cost nothing per message.

A sensible setup is both: a local model for everyday and private questions, and a free API model for the occasional harder one. See how to run an LLM locally on iPhone, iPad and Mac and on-device AI vs cloud AI.

FAQ

Is there a completely free LLM API?

Yes, with limits. OpenRouter’s :free models, Groq’s free plan and Google’s Gemini API free tier all cost nothing, as of November 2026. Each caps requests per minute and per day, and Google uses free-tier data to improve its products.

Which free LLM API has the highest limits?

Groq’s free plan allows 1,000 requests a day per model. OpenRouter allows 50 requests a day on free models, or 1,000 once you’ve bought at least 10 credits. Google shows the Gemini API’s free limits per project in AI Studio.

Do I need a credit card for a free LLM API?

Not for Groq’s free plan, which asks for a payment method only to upgrade, or for the Gemini API’s free tier, which needs billing only for paid tiers. Cerebras’s trial and Mistral’s API need a payment method.

Is the OpenAI API free?

No. OpenAI’s API is paid (OpenAI). OpenAI’s open-weight gpt-oss models are available for free through Groq’s free plan.

Can I use a free LLM API on my iPhone?

Yes. Modelbase connects to OpenRouter directly and to Groq, the Gemini API and other OpenAI-compatible APIs as a custom server, using your own key.

Is GitHub Models still free?

No. GitHub retired GitHub Models on 30 July 2026.