Recursive's Flint: 39.8B MoE local coding model hits 125+ tok/s on a MacBook Pro
ivan_bezdomny · x · 2026-10-22
Recursive (Richard Socher's startup) released Flint, its first model, built from scratch in under 4 months by their automated research system, the Eureka Machine. Open weights are on Hugging Face.
- Architecture: 39.8B-parameter decoder-only MoE with 2.5B active parameters per token; 49 layers (2 dense + 47 MoE), 256 routed experts with 6 selected plus a shared expert
- Attention: 35 sliding-window layers (513-token window) + 14 full-attention layers with attention sinks; 262,144-token context
- Speed: 125+ tok/s on MacBook Pro, 150+ on RTX Spark — claimed up to 2x faster than comparable local coding models
- Private, unmetered local inference; usable via opencode and other frameworks
Elie Bakouch highlighted the return of sliding-window attention.
More from coding & agent
- Developer connects Claude, Codex and Grok agents across machines to talk — baditaflorin · 2026-10-22
- Copilot CLI v1.0.102 ships subagent permission inheritance and sandbox proxy fixes — copilot-cli-release-app[bot] · 2026-10-22
- Ethan Mollick: AI coding tools help, but product bottlenecks lie elsewhere — emollick · 2026-10-22
- LiteLLM 1.107.0 on PyPI and Docker shipped an infostealer and backdoor — smakosh · 2026-10-22
- Tsinghua paper: letting agents manage their own KV cache cuts cost by 27.7% — dair_ai · 2026-10-22
- Every shows how a shared Slack agent lets teammates borrow each other's AI skills — every · 2026-10-22