OpenRouter currently has 17 free text models, all with model IDs ending in :free, plus a router called openrouter/free Mthat picks one for you. The biggest are NVIDIA's Nemotron 3 Ultra (550B parameters, 1M token context) and Thinking Machines' Inkling. For coding, the free picks are Poolside's Laguna S 2.1 and Cohere's North Mini Code.
Free models allow 20 requests a minute and 50 requests a day. Once you have bought $10 of credits in total, that rises to 1,000 requests a day, and the models themselves stay free.
I pulled this list from OpenRouter's public models API on October 25, 2026. The free lineup changes every few weeks, so the end of this post shows how to get the live list yourself in one command.
What are all the free models on OpenRouter right now?
Model | Model ID | Context | Input | Made for (per its model card) |
|---|---|---|---|---|
NVIDIA Nemotron 3 Ultra |
| 1M | Text | Reasoning and orchestrating agents |
NVIDIA Nemotron 3 Super |
| 262K | Text | Multi-agent work, efficient general use |
NVIDIA Nemotron 3.5 Super VL |
| 1M | Text, image, video | Agents that need to see screenshots, documents and video |
NVIDIA Nemotron 3.5 Lightning |
| 1M | Text | High-volume agent tasks |
NVIDIA Nemotron 3 Nano Omni |
| 256K | Text, audio, image, video | Perception sub-agent |
Thinking Machines Inkling |
| 1M | Text, image, audio | General use |
Thinking Machines Inkling Small |
| 1M | Text, image, audio | Lighter general use |
Poolside Laguna S 2.1 |
| 262K | Text | Coding agents |
Poolside Laguna XS 2.1 |
| 262K | Text | Smaller coding agent |
Cohere North Mini Code |
| 256K | Text | Agentic coding |
Google Gemma 4 31B |
| 262K | Text, image, video | General use, dense model |
Google Gemma 4 26B A4B |
| 262K | Text, image, video | Fast general use |
Dots3-Note Preview |
| 512K | Text, image | General use (preview) |
XingChen Xing4.0 29B A4B |
| 262K | Text | General use |
Recursive Flint |
| 262K | Text | General use, small and fast |
LiquidAI LFM2.5 2.6B |
| 65K | Text | Extraction, RAG, simple agent steps |
NVIDIA Nemotron 3.5 Content Safety |
| 128K | Text, image | Moderation only, not a chat model |
All of them except the Content Safety model support tool calling, which matters if you plan to use them with agents.

Which free OpenRouter model should you use?
Best all-rounder: Nemotron 3 Ultra. It's the biggest free model and the most used one on OpenRouter's free collection by a wide margin (6.4 trillion tokens in a week when I checked, more than four times the next model).
Coding and coding agents: Laguna S 2.1 first, North Mini Code second. Both were built specifically for agentic coding.
Screenshots, PDFs or images: Nemotron 3.5 Super VL or Gemma 4 31B.
Fast, cheap steps in a pipeline (classifying, extracting, routing): Nemotron 3.5 Lightning or LFM2.5 2.6B.
Don't care, just want a free answer: use
openrouter/freeand let OpenRouter choose.
A practical tip: free models are good enough for prototypes, scripts and learning. Before you build a product on one, run the same 10 prompts through two or three of them and compare. Model cards and parameter counts tell you what a model was built for, not how well it handles your prompts.
What are the OpenRouter free model limits?
From the OpenRouter limits docs:
Credits you have ever bought | Requests per minute | Requests per day |
|---|---|---|
Less than $10 | 20 | 50 |
$10 or more | 20 | 1,000 |
Three things the table doesn't tell you:
The $10 is a one-time threshold, not a subscription. It counts credits bought in total, and the free models don't spend those credits. So $10 once turns 50 free requests a day into 1,000 for as long as you keep the account.
50 a day is less than it sounds. A chatbot test is fine. A coding agent like Claude Code or OpenCode can make dozens of requests for one task.
Free models can be busy. At peak times a free model can return errors or slow down, because it shares capacity with everyone else using it for free.
How do you use OpenRouter free models?
Create an account at openrouter.ai. No card needed.
Go to Keys and create an API key.
Call any model ID from the table above. OpenRouter speaks the OpenAI API format.
from openai import OpenAI
import os
client = OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key=os.environ["OPENROUTER_API_KEY"],
)
reply = client.chat.completions.create(
model="nvidia/nemotron-3-ultra-550b-a55b:free",
messages=[{"role": "user", "content": "Explain MCP in two sentences"}],
)
print(reply.choices[0].message.content)To let OpenRouter pick a free model for you, set model to openrouter/free. According to OpenRouter's docs, it picks at random among free models that support what your request needs (images, tool calling, structured output), and it only falls back to other free models, so it never bills you.
Can you use OpenRouter free models in Claude Code?
Yes, with two caveats. OpenRouter has an Anthropic-compatible endpoint, so Claude Code can talk to it with these settings from OpenRouter's Claude Code guide:
export OPENROUTER_API_KEY="your-openrouter-key"
export ANTHROPIC_BASE_URL="https://openrouter.ai/api"
export ANTHROPIC_AUTH_TOKEN="$OPENROUTER_API_KEY"
export ANTHROPIC_API_KEY=""
export ANTHROPIC_DEFAULT_SONNET_MODEL="poolside/laguna-s-2.1:free"Run /logout first if you were signed in with an Anthropic account, then start claude and check /status.
The caveats: OpenRouter says Claude Code is only guaranteed to work with Anthropic's own models, so a free model may handle tools less reliably. And with 50 requests a day you will run out fast, so the $10 threshold is close to required here. If you want other ways to run Claude Code for free, I compared them in my free unlimited Claude Code guide.
Are OpenRouter free models private?
OpenRouter itself doesn't store your prompts unless you opt in. The model providers are a different story: each free model runs on a provider with its own data policy, and some providers may store prompts or train on them.
You can block those providers per request:
{
"model": "openrouter/free",
"messages": [{"role": "user", "content": "Hello"}],
"provider": { "data_collection": "deny" }
}There is also an account-wide version of this in OpenRouter's privacy settings. Expect fewer free models to be available with it on, because some free models are only offered by providers that collect data. Either way, don't send passwords, client data or anything private to a free model.
How do you check which OpenRouter models are free today?
The list changes often: new free models appear, and others quietly become paid. You can get the live list without an API key:
curl -s https://openrouter.ai/api/v1/models | python -c "import json,sys; [print(m['id']) for m in json.load(sys.stdin)['data'] if m['id'].endswith(':free')]"Or open the free models collection on OpenRouter's site. If you build something on a free model, keep the model ID in a setting, not hardcoded, so you can swap it in seconds when it disappears.
Frequently asked questions
Is OpenRouter free to use?
Creating an account and using the :free models costs nothing. Every other model is paid per token. OpenRouter adds no fee to free models.
How many free requests do you get on OpenRouter?
50 requests a day and 20 a minute on free models. After you have bought $10 of credits in total, the daily limit rises to 1,000.
Does OpenRouter need a credit card?
No. You only need a card if you want to buy credits, for example the one-time $10 that raises the free daily limit.
What is the best free model on OpenRouter?
As of October 2026, Nemotron 3 Ultra is the biggest and most used free model, and Laguna S 2.1 is the free model built specifically for coding agents. Test both on your own task, since the right answer depends on what you're building.
Why did my OpenRouter free model stop working?
Free models get removed or turned into paid models without much notice, and busy models can return errors at peak times. Check the live list with the command above and switch to another :free model or to openrouter/free.
For more free options beyond OpenRouter, see every free LLM API that still works in 2026.