Recursive launches Flint: 40B MoE local coding model built in 4 months by its automated research system
RichardSocher · x · 2026-10-22
Recursive, Richard Socher's startup, released open-weight Flint, a mixture-of-experts model for fast agentic coding on local hardware.
- 40B total parameters with only 2.5B active per token; built from scratch by a small team in under 4 months
- Leads MoE models under 50B on agentic coding benchmarks: 89.7 on SWE-Bench Pro v2, 74.2 on SWE-Bench Verified, 54.9 on Terminal-Bench 2.1
- Speed: 125+ tokens/s on an M4 Pro MacBook Pro (131 via MLX) and 150+ tokens/s on a 64GB RTX Spark — roughly 2x faster than comparable sparse models and 5x+ faster than similarly scoring dense models; fully offline and private
Flint also validates Recursive's Eureka Machine, an automated AI research system running the full loop of ideation, implementation and validation. It contributed a sparser model shape, low-bit training that shrank the model 71% with minimal quality loss, custom kernels, and inference optimizations. Weights are on Hugging Face, runnable via ollama, llama.cpp, vLLM and more.
More from coding & agent
- Skip the WordPress upgrade: have a coding agent rebuild your site in Astro — kevinkern · 2026-10-22
- Amazon blocks shopping agents like Muse, and merchants who welcome agents could win — astrange · 2026-10-22
- Claude Opus 5.5 Builds Interactive Playground Using Apple Hardware Ray Tracing — dreamwieber · 2026-10-22
- VC: portfolio startups are shipping agents for compliance, chemistry, hiring and more — atShruti · 2026-10-22
- Highlights: Hand-written WASM Syntax Highlighter Beats Shiki by Up to 298x on Large Files — cnakazawa · 2026-10-22
- Self-Evolving Agent Harnesses Overfit Without Acceptance Rules, RRSI Paper Finds — idanbeck · 2026-10-22