Recursive Debuts Flint, a Laptop Coding Model Built Largely With AI
Startup Recursive has released Flint, a free coding model that runs locally on laptops and was created with the help of an automated research system.
Recursive, an artificial intelligence startup founded last year, has released its first software creation named Flint. The company is focused on developing systems that can improve themselves, and Flint serves as an early public test of that concept. Led by CEO Richard Socher, Salesforce's former chief scientist, Recursive exited stealth mode in May having raised $650 million at a $4.65 billion valuation from investors like GV, Nvidia, and AMD Ventures.
Flint is a free, open-source coding model available on Hugging Face under the Apache 2.0 license. Because it operates directly on a user's local hardware rather than relying on external cloud servers, developers can maintain code privacy, work offline, and avoid usage fees typically charged by cloud-based AI providers.
Built With the Eureka Machine
A notable aspect of Flint's creation is the heavy involvement of Recursive's automated research system, known as the Eureka Machine. The startup states that this system contributed to nearly every phase of development—spanning pre-training, post-training, and inference—allowing a small team to build the model from scratch in under four months.
The Eureka Machine generated ideas, executed experiments, and applied the findings to determine subsequent steps. Its contributions included designing a sparser model architecture, implementing low-bit training that reduced the model's footprint by 71% without notable quality loss, and creating custom computational kernels to accelerate processing speed.
Technically, Flint utilizes a mixture-of-experts architecture. It activates only 2.5 billion of its total 40 billion parameters for any given task. This design keeps the model lightweight compared to other open alternatives, such as Meta's Muse Spark Open, which requires terabytes of memory.
In company performance evaluations on a MacBook Pro utilizing an M4 Pro chip, Flint reached speeds of roughly 131 tokens per second, outpacing similar models like Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning on the same hardware. On Nvidia's RTX Spark, it exceeded 150 tokens per second. Furthermore, on SWE-Bench Verified, it surpassed other models containing fewer than 50 billion parameters.
While early testing by journalists successfully built small applications like to-do lists and research tools, the startup noted that Flint benefits from breaking large assignments into smaller steps compared to larger paid assistants.
What it means for developers
For engineers and programmers, local models like Flint offer distinct advantages regarding data privacy and offline capability, eliminating cloud processing costs. However, developers who require managed cloud infrastructure can also try top AI models cheaply through one API at https://apixoai.online, accessing leading options like Claude, GPT, Gemini, Grok, and DeepSeek via a single key.
Recursive's long-term objective for the Eureka Machine is to achieve recursive self-improvement under human supervision, where each generated model helps the research system build subsequent iterations more rapidly.
Source: Recursive launches Flint, a laptop coding model it says was largely built by AI — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
