Recursive Flint (Free): Specs, Limits and How to Use It
Recursive Flint free: open-weight mixture-of-experts model, text input, 262K context. Specs, limits, test prompts and how to try it.
By ShilpBot Team · Published · 3 min read
Flint is an open-weight mixture-of-experts model from Recursive, with about 2.5 billion active parameters out of 40 billion in total. According to its listing it was built by Recursive's autoresearch system, called Eureka Machine, for coding agents.
The small active parameter count suggests it is designed to be quick and cheap per token, while the large total size gives it more stored knowledge than a dense 2.5B model. That combination is attractive for agents that make many small calls.
Specifications checked against the public OpenRouter catalogue on 2026-11-16. See the live model page for current limits.
Recursive Flint quick facts
| Detail | Value |
|---|---|
| Developer | Recursive |
| Model ID | recursive/flint-v1.0:free |
| Added to the catalogue | 2026-10-21 |
| Type | Open-weight mixture-of-experts model |
| Parameters | 2.5B active / 40B total |
| Context window | 262K tokens (262,144) |
| Maximum output | 235,929 tokens |
| Input types | text |
| Price | Free |
Feature support from the catalogue: Tool calling: listed; JSON / structured output: listed; Reasoning: supported, effort not adjustable.
What Recursive Flint is good for
- High-volume coding-agent calls where speed matters more than depth.
- Experiments with open-weight models you might later self-host.
- Long generation tasks, given the output cap of 235,929 tokens.
Limits of Recursive Flint
- No image input. Describe screenshots in text.
- Only a free listing appears in the catalogue, so there is no paid fallback if the free tier ends.
- A small active size can struggle with long, multi-step reasoning. Break tasks into smaller steps.
How Recursive Flint compares
- Cohere North Mini Code: It has a larger context window (262,144 against 256,000 tokens); its output cap is 235,929 tokens against 64,000; it activates 2.5B parameters per token against 3B.
- NVIDIA Nemotron 3.5 Lightning: It has a smaller context window (262,144 against 1,000,000 tokens); its output cap is 235,929 tokens against 65,536; it activates 2.5B parameters per token against 3B.
- Particle AI Particle-1: It has a smaller context window (262,144 against 1,048,576 tokens); its output cap is 235,929 tokens against 131,072; it does not accept image input.
Calling Recursive Flint from code
Use the model ID recursive/flint-v1.0:free with any OpenAI-compatible client pointed at OpenRouter:
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "recursive/flint-v1.0:free", "messages": [{"role": "user", "content": "Explain what you can do in two sentences."}]}'Prompts that show what Recursive Flint can do
- Ask it to write unit tests for a short function and then to explain which edge cases the tests miss.
- Give it a stack trace and the offending function and ask for the likely cause.
- Ask for a plan to split a large file into modules, then check the plan against the real file.
What Recursive Flint costs
The catalogue lists only a free version of this model, with no paid listing to fall back on. If the free tier changes, you will need to switch to a different model. Free pricing is set by the provider and can change at any time, so check the model page before you plan around it.
Keep reading
All free models are compared in our free AI chat models roundup. New to terms like mixture-of-experts? Start with how neural networks learn.
Frequently asked questions
What does 2.5B active out of 40B total mean?
In a mixture-of-experts model only some of the parameters are used for each token. Here about 2.5 billion are active per token, even though the full model holds about 40 billion.
Is Recursive Flint free, and what are its limits?
As of 2026-11-16 the catalogue lists it at zero price. 262K tokens (262,144) context window, up to 235,929 output tokens, model ID recursive/flint-v1.0:free. Free use can be rate limited and providers may change it.
