Recursive launches Flint: an open 40B MoE model built for local agentic coding
josh_tobin_ · x · 2026-10-22
Recursive, led by Josh Tobin, released Flint, an open-weights 40B MoE model (2B active params) designed for agentic coding running locally on laptops and consumer GPUs, with the thesis that everyday coding will move on-device.
- Runs via transformers, llama.cpp and MLX; tops 125 tokens/s on an M4 Pro MacBook Pro (131 tok/s with MLX), 156 tok/s on RTX Spark (NVFP4)
- Charts show it leading sub-50B MoE peers (Qwen3.6 35B, Gemma 4 31B, Laguna XS) on mean score across 6 agentic coding benchmarks through 128K context
- Built from scratch in under four months by a small team with an automated research system that contributed training recipes and inference kernels
- Positioning: fast, private, customizable, no API fees — not a replacement for frontier cloud models, but a fast companion for iterative making
More from coding & agent
- Resend launches Inboxes: a shared inbox for humans and their AI agents — darian314 · 2026-10-22
- How to stop Claude Code from torching your API budget overnight: gateway routing explained — Lord_Alpha01 · 2026-10-22
- Four Gemini agents beat Minecraft's Ender Dragon in 27 hours — dying 605 times — krishnan · 2026-10-22
- Dev take: AI coding frees you to think deeply — and exposes that most can't — KevinNaughtonJr · 2026-10-22
- AI-Era PRD Template: Build 5 Prototypes First, Then Write the PRD for the Survivor — aakashgupta · 2026-10-22
- Scoble builds hourly AI news pages from X lists with GrokBot, cut $6,000/mo API bill — Scobleizer · 2026-10-22