Recursive's Flint: a 40B local coding model built in 4 months, 125+ tok/s on a MacBook
jeffclune · x · 2026-10-22
Recursive (Richard Socher's startup) launches Flint, its first model: a blazing-fast 40B (2B active) local coding model built from scratch in under 4 months by their automated research system, the Eureka Machine.
- Positioned on the local-coding Pareto frontier: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark — up to 2x faster than comparable models.
- Benchmark chart on M4 Pro (48GB) shows Flint MLX/llama.cpp significantly extending the quality-vs-decode-speed frontier over Qwen3, Gemma and other baselines at 128K context.
- Open weights on Hugging Face; runs locally or free on opencode and pidotdev. Pitch: private, unmetered intelligence on your own machine.
More from Models
- HeyGen's voice models take top two spots on Artificial Analysis TTS leaderboard — HeyGen · 2026-10-22
- Base Labs open-sources mammoth-1, claims world's best decision model — philipkiely · 2026-10-22
- WIRED profiles Recursive's AI-builds-AI bet as insiders call Flint 'scary good' — _rockt · 2026-10-22
- Rumor: GLM 5.5 to Hit 2.5T-3.5T Parameters With Looped Transformer Architecture — scaling01 · 2026-10-22
- You Can Judge a New LLM by Which Benchmarks Its Team Picks for the Announcement — mariofilhoml · 2026-10-22
- Jensen Huang Keynote Demo Shows 'GPT-6 Astra High' in the Model Picker — LauraModiano · 2026-10-22