Recursive released Flint, its first model, as an open-weight coding model for developers who want agentic work on their own machine. Flint is a mixture-of-experts model with 40 billion total parameters and 2.5 billion active per token, built to investigate codebases, fix bugs, implement features, and write tests while keeping code local and avoiding API fees. Weights are on Hugging Face under Apache 2.0. Recursive says an optimized inference stack produces more than 125 tokens per second on an M4 Pro MacBook Pro and more than 150 on a 64GB NVIDIA RTX Spark at short contexts, up to twice as fast as comparable models. The launch is the first public test of the Eureka Machine, Recursive's automated research system. That system proposes ideas, implements them, tests them, and uses the results to decide what to try next. The company says it shaped nearly every part of Flint's stack, from data and training recipes to a sparser architecture, low-bit quantization, and custom kernels, which is how a small team of people and agents trained the model from scratch in under four months. Recursive reports that Flint leads mixture-of-experts models under 50 billion parameters on SWE-Bench Pro v2, Terminal-Bench 2.1, and SWE-Bench Verified, with scores of 89.7%, 54.9%, and 74.2%. It also notes that complex tasks may need more guidance than larger closed coding models. Recursive, led by CEO Richard Socher, came out of stealth in May 2026 with $650 million in funding at a $4.65 billion valuation, from backers including GV, Nvidia, and AMD Ventures. The company is applying automated research first to the AI model-building pipeline, with a longer aim of extending that loop to other sciences. Flint is the first model that process has shipped, and Recursive's stated bet is that more capable models will make the Eureka Machine a better researcher for the next one.