OpenRouter free models: the full list and the real limits

Every free model on OpenRouter right now

The table below covers the 10 general, reasoning and multimodal OpenRouter free models we checked on October 29, 2026. OpenRouter's pricing page says 25+ free models, but its models API listed 16 :free model IDs that day. No Claude, GPT or Gemini model has a :free variant.

• Nemotron 3 Ultra (55B active out of 550B parameters) and Inkling (41B active out of 975B) are the largest free models.
• Gemma 4 31B and 26B A4B both take image input on OpenRouter, with a 262,144-token context window.
• The newest additions in this group are Nemotron 3.5 Super VL (added October 15, 2026) and Xing4.0 29B A4B (added October 16, 2026).

Free coding and small models

The next table covers the rest of the :free lineup: 4 coding-agent models, 1 compact reasoning model for extraction and RAG, and 1 safety classifier.

• Poolside's Laguna S 2.1 runs 8B active out of 118B total parameters, and Recursive's Flint v1.0 (added October 21, 2026) is the newest free model on OpenRouter.
• Nemotron 3.5 Content Safety (nvidia/nemotron-3.5-content-safety:free) is a 4B guardrail model that moderates LLM inputs and outputs rather than a chat model.

OpenRouter free tier limits

OpenRouter's free tier limits apply to model IDs ending in :free (BYOK requests and exempt accounts or endpoints aside), and the daily cap is set by how many credits you have bought in total. OpenRouter's stated threshold is 10 credits ($10), but it grants the 1,000 a day cap from 9 credits to absorb rounding and top-up fees.

• Your daily quota is shared across all free models, and creating extra API keys or accounts won't raise it.
• Calling GET https://openrouter.ai/api/v1/key returns a free_model_daily_requests object showing used, limit and remaining for the current UTC day.
• An HTTP 429 error can mean you hit OpenRouter's rate limit, or that the upstream provider is at capacity.
• If your account has a negative credit balance, requests to :free models can return 402 errors until the balance is back above zero.

Who sees your prompts on free models

OpenRouter doesn't store prompts or responses unless you opt in to logging or to letting OpenRouter use them, or content is flagged as potential CSAM, though it does store metadata like token counts and latency. Each upstream provider sets its own data policy, and OpenRouter tags 10 of the 16 free endpoints as possibly training on your prompts (OpenRouter calls these tags its best knowledge, not a definitive source of third-party policies).

• All 5 NVIDIA models, both Thinking Machines Inkling models, both Poolside Laguna models and Liquid's LFM2.5 may train on prompts. OpenRouter's provider table says Poolside doesn't train, but its Laguna model pages say free usage may be used for training.
• Google AI Studio doesn't train on Gemma 4 prompts but keeps them for 55 days, while Cohere keeps North Mini Code prompts for 30 days without training.
• Xing4.0 (xingchen-agi/xing4.0-29b-a4b:free) served through Phala is the only free endpoint that neither trains on nor retains prompts.
• Turning off free-model training in your account privacy settings, or passing provider: {"data_collection": "deny"} in the request body, stops OpenRouter from routing to providers that train.

Call a free model or the free router

You can pass any :free model ID to OpenRouter's OpenAI-compatible chat endpoint, or set model to openrouter/free to let the router pick a random free model that supports what your request needs, like image input, tool calling or structured outputs. When you use openrouter/free, check the response's model field to see which model answered. OpenRouter's FAQ notes that free models have low rate limits and usually aren't suitable for production.

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openrouter/free",
    "messages": [{"role": "user", "content": "Explain rate limits in one sentence."}]
  }'

When you need Claude, GPT or Gemini models

OpenRouter's free list doesn't include any Anthropic Claude, OpenAI GPT or Google Gemini models, so calling those on OpenRouter always bills per token. Other free options are compared on our free LLM API page.

If you already pay for Claude Pro or Max, a ChatGPT plan, Google AI Pro or Ultra, or a Kimi membership, Plan2API serves that plan's models, such as claude/claude-opus-5-5, openai/gpt-6-astra, gemini/gemini-4-argon-high and kimi/k3, through an OpenAI-compatible POST /v1/chat/completions endpoint on your subscription's own usage limits, with no per-token bill. The Plan2API Free plan ($0, no card needed) covers 1 connected subscription and 50 chat requests a day, and you can read how it compares as an OpenRouter alternative.