Skip to content
NewA calmer, faster ZeroHour
StoryModel releaseinfo

Recursive releases Flint, a free 40B local coding model

3 sources · 3 articlesFirst reported Updated ()
What's new

First merged summary.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Recursive has released Flint, an Apache 2.0 mixture-of-experts coding model that runs locally with 2.5 billion active parameters.

Recursive released Flint, an open-weight mixture-of-experts model for local and agentic coding. Of 40 billion parameters, 2.5 billion are active, and the weights are on Hugging Face under Apache 2.0. The company says its Eureka Machine research system handled nearly every step, including architecture, training, quantization, and kernels, letting a small team build the model in about four months. Recursive reports strong local speed and coding-benchmark results; The Decoder says those results have not been independently verified. Flint supports llama.cpp, Ollama, LM Studio, vLLM, SGLang, and Recursive’s rec engine.

Key facts
  • Weights are free on Hugging Face under the Apache 2.0 license.
  • Recursive recommends a 256,000-token cap; the window can reach 512,000.
  • It claims about 131 tokens per second on an M4 Pro MacBook Pro.
  • An NVFP4 build is said to trail the 16-bit model by 0.7 points.
  • The startup left stealth in May after raising $650 million.
  • On Recursive’s six-test average, only Qwen3.8 27B scored higher.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

What each outlet adds

beyond the shared facts
  • TechCrunchAdds the $4.65 billion valuation, 131 tok/s figure, and six-test ranking.
  • The DecoderAdds context limits, runtime support, the NVFP4 gap, and unverified benchmarks.
  • MarkTechPostConfirms the open-weight, local-agent framing but found no further notes.
Vendors
Recursive
Products
Eureka Machinerec
Organizations
RecursiveHugging Face
AI models
Flint

Coverage

oldest first · one row per article · UTC
  1. · 4d ago TechCrunch · AIFirst report

    Recursive releases Flint, a free laptop coding model built largely by AI

    Flint activates 2.5 billion of 40 billion parameters and is free on Hugging Face under Apache 2.0.

  2. · 4d ago MarkTechPost

    Recursive releases Flint, an open-weight 40B coding model

    Flint is a mixture-of-experts model with 2.5 billion active parameters, aimed at local coding agents.

  3. · 3d ago The Decoder

    Recursive releases Flint, an open MoE coding model built by Eureka Machine

    The 40B-parameter model keeps 2.5B active per token, ships under Apache 2.0, and its benchmark lead is unverified.