Release2026-10-24
Recursive released Flint, a 39.8B-parameter mixture-of-experts coding model. It uses about 2.5B active parameters per token. Flint scores 54.9% on Terminal-Bench 2.1, 89.7% on SWE-bench Pro v2, and 74.2% on SWE-bench Verified. It supports text, reasoning, and multi-turn tool use. The model has a 262,144-token context window. Unsloth published GGUF quantizations for local use. Quantized versions range from 11.4 GB to 79.6 GB. Flint is licensed under Apache 2.0. It is designed for code investigation, bug fixing, feature implementation, and test writing.
What it unlocks
Running a 39.8B-parameter mixture-of-experts coding model locally on consumer hardware via quantized GGUF files.
- 39.8B total parameters
- 2.5B active parameters per token
- 54.9% on Terminal-Bench 2.1
- 89.7% on SWE-bench Pro v2
What you need to act on it
- a local runtime supporting GGUF files
- hardware with sufficient memory for the selected quantization
Sources