5 Sources
[1]
Recursive launches Flint, a laptop coding model it says was largely built by AI
Recursive, an AI startup founded last year, is starting to live up to its name. The company, which is trying to build AI that can improve itself, released its first model, Flint, on Wednesday. Flint is a free coding model that can run on a laptop and was built largely with the help of
[2]
This Startup Is All In on Recursive Self-Improvement. Its Latest Model Is Scary Good
For the past few days I've been testing an impressive open-weight AI coding model called Flint. It suggests that fully recursive self-improvement, or RSI -- an exciting and controversial approach that uses AI to build AI -- is getting closer. Flint is the first offering from Recursive, a company
[3]
Recursive Flint is a free laptop coding model built largely by AI
Recursive Flint runs on a MacBook and scores well on coding tests. The startup behind it wants AI that improves itself, an idea some researchers fear could slip out of human control. Recursive Superintelligence, the startup trying to build AI that improves itself, has released its first model.
[4]
Flint Could Be the First Coding AI Built via an RSI System
Flint, developed by Recursive, introduces a new approach to AI-driven programming with its "mixture of experts" architecture. Unlike traditional dense models, Flint activates only 2.5 billion of its 40 billion parameters per token, significantly reducing computational overhead while maintaining
[5]
Recursive Flint AI: Free 40 Billion Coding Model Runs Offline
Recursive has launched Flint, a free 40-billion-parameter AI coding model that runs locally on compatible laptops. Built with help from its Eureka Machine AI research system, Flint supports offline coding, offers open-source access and avoids cloud fees. AI startup Recursive launched Flint on
Share
Copy Link
Recursive unveiled Flint, a free 40-billion-parameter AI coding model that runs on laptops and was built largely by its automated research system, the Eureka Machine. The model outperforms competitors on coding benchmarks while reigniting debates about recursive self-improvement and whether AI building better AI could slip beyond human control.

Recursive launched Flint on Wednesday, introducing a free AI coding model that runs directly on laptops and was built largely by AI itself
1
. The 40-billion-parameter model represents an early public test of recursive self-improvement, where AI systems build better AI, which in turn builds even better AI2
. Recursive's automated research system, the Eureka Machine, contributed to nearly every aspect of Flint's development, allowing a small team to build the model from scratch in under four months3
. The company emerged from stealth in May with $650 million in funding at a $4.65 billion valuation, backed by GV, Nvidia, and AMD Ventures1
.Flint is available for free download on Hugging Face under the Apache 2.0 license, an open-source license that allows free use and modification
5
. Because it runs on a user's own computer rather than cloud servers, developers can use it offline, keep their code private, and avoid usage fees that AI cloud services typically charge1
. Recursive CEO Richard Socher, Salesforce's former chief scientist, leads the company's push toward AI-assisted research and ultimately aims for the Eureka Machine itself to achieve full recursive self-improvement, with human oversight and the ability to stop the process if needed2
.Flint uses a mixture-of-experts model design, meaning it activates only a small slice of itself for any given task: just 2.5 billion of its 40 billion parameters at a time
1
. Out of 256 experts, six are dynamically activated per token, along with one shared expert that remains active throughout4
. This selective activation significantly reduces computational overhead while maintaining competitive performance, making Flint light enough for an ordinary laptop and far smaller than Meta's Muse Spark Open, which uses 816 billion parameters and requires at least a terabyte of memory1
.The model incorporates sliding window attention across most layers, reducing computational costs for long-context processing
4
. Flint can handle up to 262,000 context tokens and generate outputs of up to 65,000 tokens, making it particularly well-suited for tasks like code generation and analysis where managing extensive context is crucial4
. The Eureka Machine contributed innovations including a sparser model shape, low-bit training that shrank Flint by 71% with almost no loss in quality, and custom kernels that speed up underlying computations1
.Flint delivers outstanding speed metrics that set it apart from similar models. In Recursive's tests, it generated about 131 tokens per second on a MacBook Pro with Apple's M4 Pro chip, roughly twice the 58 to 69 tokens per second that similar models managed on the same machine, including Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning
1
. On Nvidia's RTX Spark laptops, Flint topped 150 tokens per second3
.On SWE-Bench Verified, a benchmark measuring how well AI can fix real-world software bugs, Flint scored 74.2% and outperformed other models under 50 billion parameters
3
. On Recursive's own average across six tests, only one model, Alibaba's Qwen3.8 27B, scored higher than Flint1
. While Flint excels in speed and memory efficiency, it may fall short of dense models in specific accuracy benchmarks, reflecting a deliberate focus on practical usability and resource optimization4
.Wired's Will Knight tested Flint on a Mac with 48GB of memory to build several small programs, including a to-do list app and a tool that scours the web for research
1
. He found that smaller coding models often ran into errors or were slow compared to Claude or Codex, but Flint nailed his coding requests every time1
. The company acknowledged that Flint would need more guidance than big paid coding assistants and suggested breaking large tasks into smaller steps1
.Related Stories
The Eureka Machine comes up with ideas, runs experiments, and uses results to decide what to try next
1
. According to the company's launch video, it contributed to nearly every part of building Flint, from pre-training to post-training and inference1
. Socher told Wired that "nearly every aspect of the entire model pipeline was actually done and improved upon by AI"2
. The system worked on the data, training recipes, infrastructure, and code that runs the model3
.Recursive didn't specify how much time the Eureka Machine saved its researchers, but it's not alone in leaning on AI to conduct research
1
. OpenAI said last month that by mid-August, its AI agents were logging 3.1 workdays for every one worked by a human researcher1
. If Recursive's bet pays off, each new model should be able to take on longer stretches of work without human help1
. 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
.The idea of AI improving itself has become one of the most debated concepts in the industry
1
. Some researchers fear that AI improving itself could slip out of human control3
. These concerns gained wider notice in September when a researcher quit Anthropic and warned that advances in AI, including recursive self-improvement, could lead to human extinction within a few years, a concern then echoed by other experts2
. The idea of recursive self-improvement has become so controversial that several bills in Congress, including one planned by Ro Khanna, propose to ban it2
.Recursive sees these fears as more science fiction than reality and believes legislation to prevent recursive self-improvement is an unnecessary impediment to progress
2
. Socher says Flint was built using a system incapable of escaping human control and that the goal is for the Eureka Machine itself to achieve full recursive self-improvement, but this would still be overseen by humans who could stop the process if it goes awry2
. "We hope that this shows people that indeed we have all the steps in place" to build such a system, Socher told Wired2
.For now, Recursive's version is more limited: an AI research assistant overseen by people that produced a model small enough for a laptop
1
. Recursive says each new model should make the Eureka Machine a better researcher, helping it build the next one faster1
. Larger labs are moving in the same direction, with Sergey Brin stating last week that Gemini now helps build the next Gemini3
. A team from Microsoft, Carnegie Mellon University, and the University of Virginia recently published work showing how an AI coding agent can recursively improve itself by modifying its own software and model weights2
.Summarized by
Navi
[4]
[5]
14 May 2026•Startups

27 Jan 2026•Startups

13 Nov 2024•Technology
