Recursive's first model, Flint, is a 40B coder that runs on a laptop
Recursive, a new AI lab from Richard Socher, released Flint under Apache 2.0: a 40-billion-parameter mixture-of-experts coding model that runs over 125 tokens a second on a MacBook Pro.
Recursive, a new AI research lab founded by former Salesforce chief scientist Richard Socher, released its first model on 21 October: Flint, a 40-billion-parameter mixture-of-experts coding model that activates roughly 2.5 billion parameters per token.
Flint is Apache 2.0 licensed, supports context windows of up to 512,000 tokens (Recursive recommends staying within 256,000 for best results), and was built from scratch in under four months. The company says it is the top-performing model in its size class, MoE models under 50B parameters, on SWE-bench Verified, SWE-bench Pro v2 and Terminal-Bench 2.1. On an M4 Pro MacBook Pro it runs at over 125 tokens a second, and over 150 on an Nvidia RTX Spark, which Recursive says is up to twice as fast as comparable models. Ollama added it the same week, running on macOS for now.
ollama run flintWhy it matters: this is a genuinely capable coding model you can run on your own hardware under a commercially permissive licence, not just call over an API. For local-first or privacy-sensitive coding agents, it’s now a serious open-weight option rather than a research curiosity.