Flint is Recursive's open-weight coding model, a mixture-of-experts design built to run coding agents quickly on a laptop rather than in the cloud.
In plain words
Flint has 40 billion parameters but uses about 2.5 billion per token: each expert layer picks 6 of its 256 expert networks, plus one shared expert that always runs. Recursive says it writes over 125 tokens a second on an M4 Pro laptop and leads expert models under 50 billion parameters on coding tests. It is built to explore codebases, fix bugs, add features and write tests, and takes up to 512k tokens. Recursive's automated research system, the Eureka Machine, helped build it. The weights are free under Apache 2.0.
A quick example
Recursive suggests Flint for coding you would rather keep local: it works without internet, your code stays on your own computer and there are no API fees. For big jobs, it advises splitting the work into smaller steps with clear instructions.
Where you'll meet it
- Coding agents that run offline on your own computer.
- Projects whose code should not be sent to a remote server.
Related
Sources
- DocsIntroducing Flint, Recursive · read 21 Oct 2026
- DocsRecursiveSI/flint, Recursive (Hugging Face) · read 21 Oct 2026
Checked by WhatIsAI against the sources above · 21 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.