Unsloth lists GGUF builds of Recursive's Flint coding model
Original title: unsloth/flint-GGUF — new model trending #30 on Hugging Face
Flint is a 39.8B MoE model with about 2.5B active parameters, scored by Recursive at 54.9% on Terminal-Bench 2.1.
A trending Hugging Face repository, unsloth/flint-GGUF, documents Recursive's Flint mixture-of-experts model for coding agents. Flint has 39.8 billion total parameters and about 2.5 billion active per token, with text input and output, reasoning, and multi-turn tool use. Recursive reports 54.9% on Terminal-Bench 2.1, 89.7% on SWE-bench Pro v2, and 74.2% on SWE-bench Verified, noting that evaluation setups differ and some comparison scores are developer-reported.
Unconfirmed: Benchmark comparisons mix Recursive's measurements with developer-reported scores, and setups vary.
- Documented context is 262,144 tokens, including prompt and output.
- Each MoE layer has 256 routed experts, six selected per token, plus a shared expert.
- Recommended sampling is temperature 1.0, top-p 0.95, and top-k 20.
- The stated knowledge cutoff is June 2026.
- Unsloth says quantized variants are prepared separately for NVIDIA and Apple Silicon.
- AI models
- Flint
From the original article
3 min read at huggingface.cotext-generation · 7,401 downloads · 147 likes · license apache-2. 0
By unsloth · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.