Recursive Flint: a free local coding model that runs on a laptop

A free coding model that runs on the laptop you already own just got real. On October 21, Recursive released Flint, a local coding model under Apache 2.0 that it says runs at more than 125 tokens per second on an M4 Pro MacBook Pro. No API bill, no code leaving the machine. The catch is in the fine print, and the fine print is honest.

What actually happened

Per Recursive's announcement, Flint is a mixture-of-experts model with 40 billion total parameters and 2.5 billion active per token. Weights are on Hugging Face under Apache 2.0. Context goes up to 512k tokens, with 256k recommended. It ships in NVFP4 quantization, which Recursive says scores within 0.7 points of the 16-bit model.

Recursive says it leads mixture-of-experts models under 50B total parameters on SWE-Bench Verified, SWE-Bench Pro v2 and Terminal-Bench 2.1. It publishes those results only as a chart, so treat "leads" as the vendor's claim until independent runs land. It runs in llama.cpp, Ollama, LM Studio, vLLM, SGLang and Unsloth.

TechCrunch reports that Recursive's automated research system helped with almost every stage of the build. A small team shipped it from scratch in under four months. Recursive, led by former Salesforce chief scientist Richard Socher, came out of stealth in May with $650 million at a $4.65 billion valuation.

The honest part: Recursive says Flint needs more guidance than the big paid assistants and recommends breaking large tasks into smaller steps.

Why it matters for your business

Some code should never hit a vendor's server. Client contracts, pricing logic, the integration that holds your payment keys. A local coding model on a laptop handles that work with zero data leaving the building and zero per-token cost.

It is a second model, not a replacement. Use Flint for the scoped, private, repetitive work: write this test, refactor this function, explain this file. Keep the frontier model for multi-file architecture work where it still pulls ahead. Route by sensitivity and size, not by habit.

Apache 2.0 is the real headline. No usage caps, no revenue thresholds, no license to re-read before every renewal. You can build it into an internal tool and own the whole chain.

Test on your hardware first. Recursive's speeds are its own averages on a MacBook Pro M4 Pro and a 64GB RTX Spark. It does not publish a minimum memory figure. Run it on the machines your team actually uses before you plan around it.

Sources: Recursive, TechCrunch.