Recursive releases Flint, a 40B open-weight coding model that runs on laptops

Flint has 40 billion total parameters with 2.5 billion active per token, and Recursive reports 54.9% on Terminal-Bench 2.1, 89.7% on SWE-bench Pro v2 and 74.2% on SWE-bench Verified. Those are the company's own measurements, and it says the model generates more than 125 tokens per second on an M4 Pro MacBook Pro. The weights are on Hugging Face with a recommended context of 256k tokens. Recursive says a small team built the model in under four months, using its Eureka Machine to propose and test training recipes, quantization and kernels. The NVFP4 quantized version scores within 0.7 points of the 16-bit model on the company's benchmarks. Recursive is led by Richard Socher, the former Salesforce chief scientist. For builders, the release offers a coding model under a permissive license that runs on a laptop without API fees. Because the scores come from the vendor, teams should run their own evaluations on their repositories before swapping it into an agent stack.