Flint 40B: Reddit users flag a coding-focused MoE model rivaling Qwen 3.6 35B locally
soyuz-1 · reddit · 2026-11-18
While the community laments Qwen not shipping a 3.8 35B, one user points to Flint 40B, a coding-oriented MoE model claiming parity with Qwen 3.6 35B and possibly better on agentic tasks and coding, running fast in their local setup. It has become their local model of choice for speed (avoiding low quants or slow speeds that QFN would require). The poster is still testing and solicits others' experiences; no systematic benchmarks yet.
More from Models
- Asked Opus to make a video explaining consciousness and existence — "no wiser but cool" — flowersslop · 2026-11-18
- theo retests GPT-6.2 Astra: better, but still 'directionally very bad' at debugging — airesearch12 · 2026-11-18
- Claude correctly folds ubiquitin from a plain prompt, no specialized tools needed — iScienceLuvr · 2026-11-18
- GPT-6.2 Astra Tops Bug-Hunt Coding Benchmark at 62.7/105, One-Fifth Sonnet's Cost — PawelHuryn · 2026-11-18
- OpenAI's enterprise Dots reportedly to cost $50/$200/$500 per tier plus usage — and businesses are turning them off — jdjohnson · 2026-11-18
- LFM3-VL-2B scores just 7% non-hallucination rate, confusing a deadly mushroom for food — crusaderky · 2026-11-18