Recursive releases Flint, an open MoE coding model built by Eureka Machine
Original title: Recursive shows first result of its AI-driven research process with Flint
The 40B-parameter model keeps 2.5B active per token, ships under Apache 2.0, and its benchmark lead is unverified.
AI startup Recursive released Flint, a mixture-of-experts language model aimed at agentic coding on local hardware and built in four months. An internal system called Eureka Machine proposed, built, tested, and followed up experiments, including architecture, quantization, and kernels. Recursive says Flint leads several coding benchmarks among MoE models under 50 billion parameters; those results have not been independently verified. Weights are on Hugging Face under Apache 2.0, with support for llama.cpp, Ollama, LM Studio, vLLM, SGLang, and Recursive’s rec engine.
Unconfirmed: Recursive’s benchmark leadership and speed figures have not been independently verified.
- Out of 40 billion parameters, 2.5 billion are active per token.
- Recursive recommends a 256,000-token cap; the window can reach 512,000.
- On an M4 Pro it claims more than 125 tokens per second at short contexts.
- An NVFP4 model is said to sit 0.7 points behind the 16-bit version.
- The company came out of stealth in May 2026 after raising $650 million.
- Vendors
- Recursive
- Products
- Eureka Machinerec
- AI models
- Flint
From the original article
4 min read at the-decoder.com
The AI startup Recursive is unveiling Flint, its first language model, a 40-billion-parameter MoE for local coding developed by former DeepMind researchers. It was created in four months using the company's in-house AI system, "Eureka Machine," which independently tests and optimizes architectures. Recursive has yet to prove whether this approach will actually lead to self-improving AI.
By Jonathan Kemper · the-decoder.com
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.