OpenRouter Free Models Tracker
Track free OpenRouter models, compare scenario fit, and find practical options for development and assistant use.
OpenRouter Free Models Tracker
OpenRouter hosts a broad model ecosystem with many options, but the key issue is not just "there are free models", it's which free models fit this use case today.
This page is a practical tracker focused on:
- Coding and engineering-related use
- Roleplay / chat scenario prompts
- JSON and structured output use cases
- Long-context use cases
The list is generated from OpenRouter model metadata and filtered by free pricing (input and output). Data refresh can be scheduled by the project scheduler for automatic updates.
What is OpenRouter?
OpenRouter is a unified API gateway providing access to hundreds of AI models from OpenAI, Anthropic, Google, Meta, Mistral, and open-source alternatives through a single standardized interface.
Instead of integrating each provider separately, developers use one API key and one endpoint. OpenRouter handles routing, fallbacks, and billing. This makes it especially useful for:
- Comparing model quality and cost across providers in one call
- Building cost-optimized pipelines that route to free or cheap models first
- Accessing models only available through specific providers
- Experimenting with new models without separate API key provisioning
Understanding Free Models on OpenRouter
Free models on OpenRouter have both input and output pricing set to 0. This is not a permanent guarantee — providers may change pricing, rate limits, or retire free tiers at any time.
Why Free Models Matter for Real Projects
Free models are not just for hobbyists. Many production scenarios benefit from free-tier routing:
- Development and testing: Run CI pipelines and evaluation scripts without burning token budgets.
- Long-tail queries: Route low-value or repetitive queries to free models, reserving paid capacity for high-stakes requests.
- Fallback chains: Configure a primary paid model with a free fallback for rate-limit or downtime scenarios.
- Research and benchmarking: Compare multiple models on the same dataset without cost constraints.
Rate Limits and Fair Use
Free models typically carry rate limits — requests per minute, tokens per day, or concurrent connections. These limits vary by provider and model. Verify limits directly on the OpenRouter model page before building a production dependency.
How to Access These Models
All free models are accessible through the standard OpenRouter API. The base URL is https://openrouter.ai/api/v1 and the interface is compatible with the OpenAI SDK.
Basic Python setup:
from openai import OpenAI
client = OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key="YOUR_OPENROUTER_API_KEY",
)
response = client.chat.completions.create(
model="model-id-from-table-below",
messages=[{"role": "user", "content": "Hello"}],
)
Filter free models via CLI:
curl https://openrouter.ai/api/v1/models | \
jq '[.data[] | select(.pricing.prompt == "0" and .pricing.completion == "0")]'
Choosing the Right Free Model
For Coding Tasks
Look for high context length, tool-calling support, and JSON mode. Key benchmarks: HumanEval, MBPP, LiveCodeBench. Longer context lets the model see more of your codebase at once.
For JSON and Structured Output
Prioritize models that explicitly support JSON mode or tool calling. Test with complex nested schemas — some models produce syntactically valid but semantically incorrect JSON at depth.
For Long-Context Tasks
Context length in the table is the maximum supported input. Effective performance often degrades after 50–70% of the advertised window. Test with your actual document lengths before committing.
For Roleplay and Chat
Instruction-following quality and personality consistency matter most. Models trained with RLHF or DPO tend to produce more natural conversational responses.
Tips for Production Use
- Pin model versions — use the full model ID including version suffix to avoid silent model swaps.
- Implement fallbacks — free models can become rate-limited; build a fallback chain in your routing logic.
- Cache responses — for repeated identical queries, application-layer caching eliminates redundant API calls.
- Use streaming — streaming reduces time-to-first-token and improves perceived performance.
- Monitor usage — even free models count toward your OpenRouter dashboard; track to spot unexpected spikes.
Related Resources
- OpenRouter documentation — full API reference and model metadata schema
- OpenRouter model rankings — community-voted quality scores per use case
- MCP Hub — browse MCP servers that integrate with OpenRouter for agent workflows
Last update
2026-08-26T20:10:04.679Z
Coding
| Model | Provider | Context Length | Price (prompt/completion) | Note |
|---|---|---|---|---|
| Thinking Machines: Inkling (free) | Unknown | 1048576 | 0/0 | Inkling is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 41B active parameters out |
| MiniMax: MiniMax M3 (free) | Unknown | 1048576 | 0/0 | MiniMax-M3 is a multimodal foundation model from MiniMax. It supports text, image, and video inputs with text output, a |
| Poolside: Laguna S 2.1 (free) | Unknown | 262144 | 0/0 | Laguna S 2.1 is the latest coding agent model from Poolside. Laguna S 2.1 is a 118B total para |
| Poolside: Laguna XS 2.1 (free) | Unknown | 262144 | 0/0 | Laguna XS 2.1 is the latest coding agent model in the 33B-A3B category from Poolside and a step |
| Cohere: North Mini Code (free) | Unknown | 256000 | 0/0 | North Mini Code is Cohere's first agentic coding model and the debut of its North family. A sparse mixture-of-experts mo |
| LiquidAI: LFM2.5-2.6B (free) | Unknown | 65536 | 0/0 | LFM2.5-2.6B is a compact reasoning model from Liquid AI. It is suited for agent workflows, data extraction, RAG, and lon |
JSON
| Model | Provider | Context Length | Price (prompt/completion) | Note |
|---|---|---|---|---|
| Google: Gemma 4 31B (free) | Unknown | 262144 | 0/0 | Gemma 4 31B Instruct is Google DeepMind's 30.7B dense multimodal model supporting text and image input with text output. |
| NVIDIA: Nemotron 3 Nano Omni (free) | Unknown | 256000 | 0/0 | NVIDIA Nemotron™ 3 Nano Omni is a 30B-A3B open multimodal model designed to function as a perception and context sub-age |
Long Context
| Model | Provider | Context Length | Price (prompt/completion) | Note |
|---|---|---|---|---|
| Thinking Machines: Inkling (free) | Unknown | 1048576 | 0/0 | Inkling is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 41B active parameters out |
| MiniMax: MiniMax M3 (free) | Unknown | 1048576 | 0/0 | MiniMax-M3 is a multimodal foundation model from MiniMax. It supports text, image, and video inputs with text output, a |
| NVIDIA: Nemotron 3 Ultra (free) | Unknown | 1000000 | 0/0 | NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters ou |
| Google: Gemma 4 31B (free) | Unknown | 262144 | 0/0 | Gemma 4 31B Instruct is Google DeepMind's 30.7B dense multimodal model supporting text and image input with text output. |
| Z.ai: GLM 5.2 (free) | Unknown | 256000 | 0/0 | GLM 5.2 is a large-scale reasoning model from Z.ai. It supports text input and output with a 1M-token context window, an |
| NVIDIA: Nemotron 3 Nano Omni (free) | Unknown | 256000 | 0/0 | NVIDIA Nemotron™ 3 Nano Omni is a 30B-A3B open multimodal model designed to function as a perception and context sub-age |
| LiquidAI: LFM2.5-2.6B (free) | Unknown | 65536 | 0/0 | LFM2.5-2.6B is a compact reasoning model from Liquid AI. It is suited for agent workflows, data extraction, RAG, and lon |