Introduction to Flint
Recursive, an AI startup founded last year, has released its first model, Flint. Flint is a free coding model that can run on a laptop and was built largely with the help of artificial intelligence.
Recursive’s automated research system, which it calls the Eureka Machine, helped with nearly every step of Flint’s development. That allowed a small team to build the model from scratch in under four months.
Flint serves as an early public test of recursive self-improvement, exploring whether AI can build better AI. While some researchers express concern that self-improving systems could slip out of human control, Recursive’s current implementation operates as a human-overseen research assistant.
The company emerged from stealth with $650 million in funding at a $4.65 billion valuation from backers including GV, Nvidia, and AMD Ventures. Recursive is led by CEO Richard Socher, formerly the chief scientist at Salesforce.
Performance and Local Execution
Because Flint executes locally on a user’s computer rather than cloud servers, developers can use it offline while maintaining code privacy and avoiding cloud subscription fees. The model is free to download on Hugging Face under the Apache 2.0 license.
Flint is a mixture-of-experts model activating only 2.5 billion of its 40 billion parameters at a time. This design keeps the hardware footprint small enough for standard laptops, contrasting with larger open models that require extensive memory configurations.
In testing on a MacBook Pro with Apple’s M4 Pro chip, the model generated approximately 131 tokens per second, outpacing several comparable models tested on the same hardware.
On SWE-Bench Verified, a benchmark evaluating bug-fixing capabilities, Flint outperformed other models under 50 billion parameters. Across Recursive's internal test averages, only one model scored higher.
While the company notes that Flint requires structured guidance for complex tasks, early third-party tests indicate the model handles defined programming requests effectively.
The Eureka Machine
The Eureka Machine generates ideas, executes experiments, and analyzes outcomes to determine subsequent steps. According to the company, it contributed to development phases ranging from pre-training to post-training and inference.
Recursive stated that Eureka's contributions include designing a sparser model architecture, implementing low-bit training that reduced the model size significantly, and developing custom computational kernels.
The Broader AI Research Trend
Recursive is part of a broader industry shift toward utilizing AI agents to accelerate research and development workflows, joining other major AI labs leveraging automated systems for operational tasks.
The company’s broader infrastructure strategy emphasizes expanding agent capacity to handle increasingly complex multi-step tasks with minimal manual intervention.
The long-term objective for the Eureka Machine is to achieve full recursive self-improvement under human oversight, with each successive iteration designed to accelerate the creation of subsequent models.
