Recursive launched Flint on October 21, 2026. Much of the research behind this open-weight coding model was conducted with the help of AI. The entire model was created with the assistance of a small team in less than four months. The question is whether this approach creates a chance for small laboratories to compete with big players without huge investment.
Flint uses a mixture-of-experts design. It activates only a small part of the model at a time, helping it run efficiently. Its model card lists 39.8 billion parameters, with about 2.5 billion active per token. The Apache 2.0 license allows commercial use, and developers can run Flint locally.
What Recursive says the Eureka Machine built
Recursive credits its Eureka Machine with suggesting, coding and testing changes to Flint’s architecture, training and inference software. CEO Richard Socher told WIRED that AI improved “nearly every aspect of the entire model pipeline.”
That is central to Recursive’s goal. It wants AI systems to help build better successors. Socher maintains that people can still stop the research process.
Benchmarks from the company, a hands-on from WIRED
In tests conducted by Recursive, Flint outperformed other mixture-of-experts models having less than 50 billion parameters on SWE-Bench Pro v2, Terminal-Bench 2.1 and SWE-Bench Verified. The company also reported speeds above 125 tokens per second on an M4 Pro MacBook Pro and 150 on an NVIDIA RTX Spark at short context lengths.
One important consideration: these are company numbers. Independent evaluation is still limited. WIRED’s Will Knight tested Flint himself and found it to be quite adept, even able to create a small website. However, some successful coding tasks do not indicate how it compares with other models.
OpenHands conducted tests on Flint prior to its launch and provided developers access to it on the day of its roll-out. Although this indicates Flint is capable of performing real-world coding tasks, OpenHands hasn’t conducted an independent assessment of Recursive’s claims regarding their benchmarking.
The $410 million question about cheap AI
Creating a model that requires fewer people is not a guarantee that it will be cheap. Recursive has not revealed how much it costs to build Flint.
On July 28, 2026, the company entered in a $410 million deal with AWS. Previously it raised $650 million. And none of these amounts show the cost of building Flint itself.
Epoch AI’s research indicates why expenses stay high. Final training runs made up 9.6% of OpenAI’s research-compute expenditures during the year 2024. The majority of the funding was distributed to other research-related actions.
AI agents can take over some jobs that used to be handled by researchers. However, running experiments and testing new ideas still require a big amount of computing power.
Open weights, local hardware, and global competition
Meanwhile, using AI is becoming much cheaper. Epoch AI estimates that the cost of achieving comparable performance has fallen about 47% per quarter since 2023.
On September 22, 2026, OpenAI introduced GPT-6 Sol and Luna at prices at least 50% below their predecessors, as Cryptopolitan reported.
Flint had nothing to do with those price changes. Still, it gives developers another option: a coding model they can run on their own hardware without relying on paid cloud services.
The benefits could reach beyond established AI markets. Stanford’s 2026 AI Index found that open-source development is helping more countries participate in AI innovation.
For now, Flint shows what a small team can accomplish with AI-assisted research. Whether the same approach can make model development affordable for smaller labs remains an open question.
If you're reading this, you’re already ahead. Stay there with our newsletter.