Flint - The First RSI Coding Model?
In this video, I look at an interesting model from Recursive AI called Flint, which is their first public attempt at building a very fast mixture-of-experts model specifically for taking on the Qwen3.8 and similar coding models. They are using an RSI system to do this. Blog: https://recursive.com/updates/introducing-recursive-flint HF: https://huggingface.co/RecursiveSI/flint NVFP4: https://huggingface.co/RecursiveSI/flint-nvfp4-gguf 2nd Channel: https://www.youtube.com/channel/UCUF7WtR4JpVbQqcQGrLeADw Twitter: https://x.com/Sam_Witteveen 🕵️ Interested in building LLM Agents? Fill out the form below Building LLM Agents Form: https://drp.li/dIMes 👨💻Github: https://github.com/samwit/llm-tutorials ⏱️Time Stamps: 0:00 Intro 0:33 Flint 1:04 Built by the Eureka Machine in under 4 months 1:34 Why it's fast 3:02 Open weights: four official versions 3:22 Benchmarks vs Qwen 3.6 and Laguna XS 3:56 Flint vs Qwen 3.8 27B 4:22 The speed vs score chart 4:34 How to run it: Ollama, LM Studio, llama.cpp, vLLM 5:48 Who's behind Recursive 8:01 What the Eureka Machine actually is 10:45 How it built Flint: architecture search 13:08 Demo: NVFP4 on an RTX PRO 6000 16:13 Agent tasks and tool use 17:10 Which GPUs can run Flint 18:02 Takeaways
