AI Briefingflint: 40B MoE Coding Agent with 2.5B Active Parameters per Token
RecursiveSI/flint
About the project
Flint, released by RecursiveSI, is a language model specialized in codebase exploration, bug fixing, feature implementation, and test writing. It adopts a mixture-of-experts structure, activating approximately 2.5B of its total 39.8B parameters per token. It supports text input and output, reasoning, and multi-turn tool calling.
Performance metrics focus on coding agent tasks. It recorded 74.2% on SWE-bench Verified, 89.7% on SWE-bench Pro v2, and 54.9% on Terminal-Bench 2.1. It features a 262,144-token context window and a 150,144-token vocabulary, with a knowledge cutoff of June 2026.
Hardware requirements are a single B300 or H200 GPU. The repository includes a BF16 checkpoint and an SGLang serving kit, enabling immediate deployment via Docker. Quantized versions compatible with llama.cpp, Ollama, and LM Studio are also separately available.
Released under the Apache-2.0 license, it can be used in commercial environments. For handling complex code issues, an output version of 32,768 tokens or more is recommended, with settings optimized at temperature 1.0, top-p 0.95, and top-k 20.
RecursiveSI/flint
The original page has no description.
text-generation
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.