Recursive ships Flint, a laptop coding model built by AI — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Recursive released Flint, a free open-source coding model that runs on a laptop and was built largely by its 'Eureka Machine' AI research system in under four months.
- Flint is a mixture-of-experts model activating 2.5 billion of its 40 billion parameters per task — far smaller than Meta's 816-billion-parameter Muse Spark Open, which needs more than a terabyte of memory.
- Flint hit ~131 tokens/second on a MacBook Pro M4 Pro, topped 150 tokens/sec on Nvidia's RTX Spark, and led models under 50B parameters on SWE-Bench Verified, with only Alibaba's Qwen3.8 27B scoring higher on Recursive's six-test average.
- Richard Socher, Salesforce's former chief scientist, leads Recursive, which exited stealth in May with $650 million at a $4.65 billion valuation from GV, Nvidia, and AMD Ventures.
- The Eureka Machine contributed a sparser model shape, low-bit training that shrank Flint by 71% with minimal quality loss, and custom kernels for inference speedups.
- OpenAI said its AI agents were logging 3.1 workdays for every one human researcher by mid-August, one of several data points backing Recursive's bet that AI-built AI can compound.
Why it matters: For developers, Flint's on-device execution eliminates per-token cloud fees and code-exposure risk, and its sub-50B benchmark wins undercut the premise that competitive coding models need Meta-scale 816B-parameter footprints — a counter to the idea that coding AI has to live in someone else's data center.
Ask SkimNews
