Recursive launches Flint coding model for laptop AI development

Traders Union
All News
Business & Wealth
Companies
Flint coding model for AI

Highlights

Recursive launched Flint, a free coding model with 2.5 billion active parameters, designed for offline use on laptops, available on Hugging Face under Apache 2.0.
Flint achieved 131 tokens per second on an Apple M4 Pro MacBook, about twice the speed of comparable models like Google's Gemma 4 and Nvidia’s Nemotron 3.5 Lightning.
Recursive exited stealth in May with $650 million funding at a $4.65 billion valuation, emphasizing AI agent capacity over headcount through a $410 million AWS compute deal.

Flint launch and development approach

As reported by TechCrunch, Recursive says its automated research system, the Eureka Machine, contributed to nearly every stage of Flint’s development, helping a small team build the model from scratch in under four months. The startup, founded last year, positions Flint as an early public test of recursive self-improvement, the idea that AI can build better AI that then helps create stronger successors.

Recursive says Flint is a free coding model designed to run locally on a user’s computer rather than through cloud infrastructure. The company says that setup lets developers work offline, keep code private and avoid usage-based charges tied to text processing. Flint is available to download on Hugging Face under the Apache 2.0 open-source license.

The model uses a mixture-of-experts architecture, activating 2.5 billion of its 40 billion parameters for a given task. Recursive says that makes Flint light enough for an ordinary laptop and materially smaller than some other open models built for local use. According to the company’s launch video, Eureka helped shape Flint’s pre-training, post-training and inference, including a sparser model design, low-bit training that reduced its size by 71% with little quality loss, and custom kernels to speed computation.

Performance claims and industry implications

Recursive says Flint generated about 131 tokens per second in its tests on a MacBook Pro using Apple’s M4 Pro chip, roughly double the 58 to 69 tokens per second it says comparable models achieved on the same machine, including Google’s Gemma 4 and Nvidia’s Nemotron 3.5 Lightning. On Nvidia’s RTX Spark, the company says Flint exceeded 150 tokens per second.

On SWE-Bench Verified, which measures how well AI models fix real-world software bugs, Recursive says Flint outperformed other models under 50 billion parameters. Across six tests averaged by the company, only Alibaba’s Qwen3.8 27B scored higher than Flint. Recursive also acknowledges that Flint needs more user guidance than large paid coding assistants and suggests breaking bigger jobs into smaller steps.

The launch adds to a broader industry push to use AI systems in research and model development. Recursive emerged from stealth in May with $650 million in funding at a $4.65 billion valuation from backers including GV, Nvidia and AMD Ventures, and the company previously said its $410 million compute deal with Amazon Web Services in June was focused less on headcount and more on expanding AI agent capacity. CEO Richard Socher says the longer-term goal is full recursive self-improvement with humans supervising the process and able to stop it if needed.

Our earlier analysis of the Nasdaq 100’s volatility linked the index’s sharp swings to worries about AI supply chains and higher financing costs weighing on technology stocks. We noted that, despite a rebound, the index remained below key moving averages, with mixed momentum signals and a focus on specific resistance and support levels for the next move.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.