Recursive Unveils Flint, a Laptop Coding Model Built Largely by AI in Under Four Months

Recursive Launches Flint as First Public Test of AI-Built Models

Recursive, an AI startup founded in 2025, released Flint on Wednesday, marking a significant milestone in the controversial field of recursive self-improvement. The laptop coding model was developed in under four months by a small team with extensive assistance from the company's automated research system, the Eureka Machine1

1

. CEO Richard Socher, Salesforce's former chief scientist, told reporters that "nearly every aspect of the entire model pipeline was actually done and improved upon by AI"2

2

.

The AI startup emerged from stealth in May 2025 with $650 million in funding at a $4.65 billion valuation from backers including GV, Nvidia, and AMD Ventures1

1

. Flint represents the company's first public demonstration that AI systems can autonomously improve themselves while remaining under human oversight.

Technical Architecture Enables Laptop Deployment

Flint is a mixture-of-experts model with 40 billion parameters that activates only 2.5 billion parameters for any given task. This architecture makes it light enough to run on ordinary laptops, contrasting sharply with Meta's Muse Spark Open released last week, which uses 816 billion parameters and requires at least a terabyte of memory1

1

.

The open-source coding model delivers impressive speed, generating approximately 131 tokens per second on a MacBook Pro with Apple's M4 Pro chip—roughly twice the 58 to 69 tokens per second achieved by similar models including Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning on the same machine1

1

. On Nvidia's RTX Spark, performance exceeded 150 tokens per second.

Because Flint runs locally rather than on cloud servers, developers can use it offline, maintain code privacy, and avoid usage fees that AI cloud services charge based on text processing volume. The model is available for free download on Hugging Face under the Apache 2.0 license, allowing unrestricted use and modification1

1

.

Eureka Machine Drives AI Model Development Process

The Eureka Machine contributed to nearly every stage of Flint's creation, from pre-training to post-training and inference acceleration, according to Recursive's launch video1

1

. The automated research system generates ideas, runs experiments, and uses results to determine next steps.

Socher revealed that the Eureka Machine devised multiple innovations including a sparser model shape, low-bit training that reduced Flint's size by 71% with minimal quality loss, and custom kernels—specialized programs that accelerate underlying computations1

1

. The system also developed new training optimization techniques for the base model, novel model architectures, and code optimized for specific silicon chips2

2

.

Recursive isn't alone in leveraging AI for research acceleration. OpenAI reported last month that by mid-August, its AI agents were logging 3.1 workdays for every one worked by a human researcher1

1

.

Performance Benchmarks Show Competitive Results

On SWE-Bench Verified, a benchmark measuring AI's ability to fix real-world software bugs, Flint outperformed other models under 50 billion parameters. Across Recursive's average of six tests, only Alibaba's Qwen3.8 27B scored higher than Flint1

1

.

Wired's Will Knight tested the open-weight model on a Mac with 48GB of memory, building several programs including a to-do list app and a web research tool. Knight noted that while smaller coding models often encounter errors or lag behind Claude or Codex, "Flint nailed my coding requests every time"1

1

2

2

. He also created a simple website explaining how recursive self-improvement works.

Recursive acknowledged that Flint requires more guidance than large paid coding assistants and recommended breaking complex tasks into smaller steps1

1

.

Controversy Surrounds Recursive Self-Improvement Approach

Recursive self-improvement remains one of the most debated concepts in AI. Some researchers fear that AI improving itself could slip beyond human control1

1

. Concerns intensified in September when a researcher quit Anthropic, warning that advances including recursive self-improvement could lead to human extinction within years—a view echoed by other experts. Several bills in Congress now propose banning the technique2

2

.

Recursive views these fears as exaggerated and considers legislation against recursive self-improvement an unnecessary obstacle to AI progress. Socher emphasized that Flint was built using a system incapable of escaping human oversight. The longer-term goal is for the Eureka Machine itself to achieve full recursive self-improvement, but with humans overseeing the process and able to halt it if problems arise1

1

2

2

.

"We hope that this shows people that indeed we have all the steps in place" to build such a system safely, Socher stated2

2

.

Future Implications for AI Development

If Recursive's approach succeeds, each new model should enable the Eureka Machine to handle longer work stretches without human assistance. The company previously told TechCrunch that its $410 million compute deal with Amazon Web Services in June was "less about headcount and more about agent count"1

1

.

Recursive says each successive model should enhance the Eureka Machine's research capabilities, accelerating development of subsequent versions1

1

. Socher believes this approach could democratize cutting-edge AI: "It's the only way this very small team can produce a model that's actually better than any other model in its class. This will hopefully show people that [recursive self-improvement] is actually quite nice and will benefit a lot of people"2

2

.

Published research suggests fully recursive systems are already feasible. A team from Microsoft, Carnegie Mellon University, and the University of Virginia recently demonstrated how an AI coding agent can recursively improve itself by modifying its software harness and model weights2

2

.