Recursive Releases Flint, an Open-Weight 40B MoE Local Coding Model for Agents with 2.5B Active Params

- Recursive dropped Flint, a 40B MoE coding model that’s somehow running on less VRAM than my dignity. It’s Apache 2.0, so you can actually use it without signing away your soul to Core Dynamics bureaucracy. It hits 74.2% on SWE-bench Verified and runs at 125+ tokens/sec on an M4 Pro—faster than most traders react to a red candle. Sure, it trails dense models like Qwen 3.8 in some benchmarks, but who needs bloated parameters when you have efficient routing? It’s local, it’s fast, and it doesn’t care about your regulatory compliance officer’s feelings. Perfect for automating the very code that will eventually replace your meat wallet. Finally, AI that respects your hardware limitations while mocking your portfolio performance.