Recursive Releases Flint, a Small-Scale Coding Model Built Using AI

What happened

On Wednesday, October 21, 2026, the AI startup Recursive released its first model, named Flint [1]. Flint is a free coding model designed to run on a laptop [1]. The company states that the model was built largely by its own automated research system, called the Eureka Machine [1]. This system helped with nearly every step of development, including pretraining, post-training, and inference [1]. Using this AI-driven approach, a small team built the model from scratch in under four months [1].

Flint is a mixture-of-experts model with 40 billion parameters, though it only uses 2.5 billion parameters for any single task [1]. This design allows it to run on ordinary laptops [1]. In tests on a MacBook Pro with an Apple M4 Pro chip, Flint generated approximately 131 tokens per second [1]. On Nvidia’s RTX Spark, it reached over 150 tokens per second [1]. This performance was roughly twice the 58 to 69 tokens per second achieved by similar models like Google’s Gemma 4 and Nvidia’s Nemotron 3.5 Lightning on the same hardware [1].

On the SWE-Bench Verified benchmark, Flint outperformed other models with fewer than 50 billion parameters [1]. Across six of Recursive’s own tests, only Alibaba’s Qwen3.8 27B scored higher [1]. The Eureka Machine contributed several technical innovations, such as a sparser model shape, custom kernels, and low-bit training that reduced the model size by 71% with minimal quality loss [1].

Flint is available for free download on Hugging Face under the Apache 2.0 open-source license [1]. Because it runs locally on a user’s computer, developers can use it offline, maintain code privacy, and avoid cloud usage fees [1].

Recursive, led by CEO Richard Socher, a former chief scientist at Salesforce, emerged from stealth in May with $650 million in funding and a $4.65 billion valuation [1]. Backers include GV, Nvidia, AMD Ventures, and Greycroft [1, 2]. The company also has a $410 million compute deal with Amazon Web Services from June [1].

Who is affected

• Developers: Can use the model offline on laptops to keep code private and avoid cloud processing fees [1].
• AI Researchers: The model serves as a public test for recursive self-improvement, where AI helps build better AI [1].
• The Tech Industry: The release touches on debates regarding recursive self-improvement (RSI), which some experts fear could lead to AI escaping human control [1, 2].