Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro
RichardSocher · x · 2026-10-27
A developer benchmarked Recursive's local model Flint (int4) against Qwen3.6 35B A3B on his Mac mini M4 Pro (48 GB) using the same 52-task local screen with no cached prompts. Quality tied at 50/52, but speed differed sharply: median task time 2.8s vs 10.1s, and output throughput 122.6 vs 91.5 tok/s on 1K prompts (90.9 vs 73.9 at 20K), roughly confirming Recursive's claimed 125+ tok/s. Qwen reads prompts 10-19% faster and peaks at 18.6 GB vs Flint's 25.5 GB, while Flint cold-loads in 9s vs 30s. Caveat: Flint always thinks before answering, and once burned an entire 8K output budget on thinking without producing an answer.
More from Infra
- Lam Research: weekly token usage up ~250x in 18 months — Beth_Kindig · 2026-10-27
- Encrypted GPT-2 speedrun challenges FHE inference: best is 3.031 s/token, goal sub-1s — tensorqt · 2026-10-27
- Open-source Latch compiles PyTorch into training megakernels, 2.75x faster on H100 — sguada · 2026-10-27
- NVIDIA Claims HGX Rubin NVL8 Delivers 10x More Tokens per Megawatt Than B200 — thursdai_pod · 2026-10-27
- Together lets multiple LoRA post-training experiments share the same GPU capacity — togethercompute · 2026-10-27
- Commoditization dilemma: either human labor or GPUs become the commodity — _AustinCalvert_ · 2026-10-27