Recursive Flint: a 40B/2.5B-active coding model that beats Qwen3.6-35B-A3B on TerminalBench and SWE-bench Pro
returnity · reddit · 2026-10-22
A Reddit user reviews Recursive's newly released Flint model, calling it the long-awaited Qwen3.5 35B alternative for coding agents.
- Architecture: 40B params with only 2.5B active, 3:1 SWA-to-full attention ratio, 262k context
- Results: decisively beats Qwen3.6-35B-A3B on TerminalBench 2.1 and SWE-bench Pro v2, ties or better elsewhere
- Positioning: aimed at local coding-agent workflows for the GPU-poor
WIRED covered the release as "scary good"; a llama.cpp support PR is already up, and the poster plans to publish hands-on benchmarks in the coming days.
More from coding & agent
- Node.js ships security fixes across 22.x/24.x/26.x, worst bug allows RCE — dotey · 2026-10-22
- Research Progress Graph: modeling authentic research trajectories, 35 agents found incomplete — ZhiyuChen4 · 2026-10-22
- Longshot Studio 0.15 adds lip sync and face refinement, free on your own ComfyUI — R34vspec · 2026-10-22
- Dev builds Scratch-style MCP that feeds coding agents one task at a time to stop plan drift — NoTokenDev · 2026-10-22
- Frontier models spark a comeback of quirky game-style personal homepages — HamelHusain · 2026-10-22
- FOSS stack with DeepSeek beats paid ChatGPT plan in 2-month agent reliability test — layer4down · 2026-10-22