Recursive's 40B Flint Hits ~200 tok/s on M3/M5 Ultra via MLX-VLM with Fused Metal Kernels
timshi_ai · x · 2026-10-24
Prince Canuma showed Recursive's 40B Flint model running at nearly 200 tok/s on Apple M3 and M5 Ultra using the MLX-VLM dev branch.
jtdavies, who benchmarked the model earlier, found the unmerged mlx-vlm PR 2569 (fused Metal kernels) made a stark difference — up to 30% improvement — and praised Flint as an agentic model that holds its own against the leaders.
More from Infra
- Cloudflare launches Origin Range Requests, cutting R2 p99 TTFB by half — Cloudflare Blog · 2026-10-24
- Tencent's Hy-Nomas world model needs at least 24GB VRAM, dev complains — cocktailpeanut · 2026-10-24
- kapitsa: hand-written Metal inference engine hits 140 tok/s on Qwen3.8-27B for local OpenAI/Anthropic APIs — DotaMate · 2026-10-24
- Humansand Speeds Up NVFP4 Decode Kernel by Up to 10% While Improving Accuracy — noahdgoodman · 2026-10-24
- Cloudflare launches post-quantum GeoKey Manager for geo-fenced key decryption — Cloudflare Blog · 2026-10-24
- LiteLLM breach lesson: AI gateway logs are a secrets store — alvelda · 2026-10-24