Recursive Releases Flint: Open-Weight 40B MoE Coding Model with 2.5B Active Parameters

Recursive has released Flint, an open-weight 40-billion-parameter Mixture-of-Experts coding model designed specifically for agentic coding workflows, with only 2.5 billion parameters active per forward pass. The MoE architecture means Flint achieves strong coding performance while keeping inference compute and memory requirements comparable to a much smaller dense model, making it practical to run locally. It is explicitly designed for use in coding agents—not just autocomplete—meaning it is optimized for multi-step task execution, code editing across files, and tool use. For developers who need a capable, locally-runnable coding model without sending code to an external API, Flint is a notable new option in a space dominated by proprietary models. The open-weight release also allows fine-tuning on proprietary codebases, which is a significant advantage for enterprise and security-conscious teams.