What's your opinion on Flint?
retained text
Flint is Recursive’s mixture-of-experts language model for coding agents, with 39.8 billion total parameters and approximately 2.5 billion active per token. I have 32 GB VRAM and 32 GB RAM. With llama.cpp I can run Qwen 3.8 27B at Q6, but I can run Flint at Q8 with experts offloaded to CPU. I get about 40 t/s in both the model. Actually, Flint gives 1-2 token more per second. What are your opinions on it?
document fetched 2026-10-29 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xi00kn/whats_your_opinion_on_flint