Recursive AI's Flint: open-weights local coding model hits 125+ tok/s on MacBook Pro
josh_tobin_ · x · 2026-10-22
- Recursive AI (@RecursiveSI), Richard Socher's new venture, released its first model Flint: a fast local coding model hitting 125+ tokens/sec on a MacBook Pro and 150+ on RTX Spark — up to 2x faster than comparable models — built from scratch in under 4 months by their automated research system, the Eureka Machine.
- Team member anhng8 describes building RSI loop(s) that produced a best-in-class local coding model: simple examples solved in under 10 minutes, "0 endpoints, 0$, full CoTs," with the promo video auto-generated by Fable from the artifacts.
- Open weights are on Hugging Face; run locally or try free on opencode or pidot.dev, pitched as private, unmetered intelligence on your own machine.
More from coding & agent
- Inside T800: Contrarian Reviews Handled by a Filter Agent Before Dispatch — lucasmeijer · 2026-10-22
- The "Seed to Scope" path for AI adoption inside companies: from one laptop to beating the team baseline — curious_vii · 2026-10-22
- Agents Search More Than Ever: Why Retrieval Quality Is Now the Bottleneck — Nils_Reimers · 2026-10-22
- Dev Leaves Claude Max Running Overnight in Two Terminals, Racks Up $2k+ Bill — Kuprel · 2026-10-22
- OpenClaw Dogfoods in Public: Building Features on X Using Its Own Agent, PR Included — vincent_koc · 2026-10-22
- AgentsInTheCloud: open-source tool to run cloud coding agents on your own server — lucasmeijer · 2026-10-22