Recursive has released Flint, a free open-weights coding model that runs locally on consumer hardware.
Built largely using the automated Eureka Machine research system in under four months, Flint uses a mixture-of-experts architecture with 2.5 billion active parameters from 40 billion total.
Available on Hugging Face under an Apache 2.0 license, it enables offline use and code privacy.
Flint was developed by a small team in less than four months, with significant input from the company’s automated research system, known as the Eureka Machine, which assisted in various stages, including initial training, inference, and custom functionality creation.
Features and function
Flint runs directly on the user's device, allowing offline use, code protection, and avoiding charges to remote cloud services. It is available for free on the Hugging Face platform under the Apache 2.0 license.
The model uses 2,5 billion active parameters out of a total of 40 billion. In tests on a MacBook Pro laptop with an M4 Pro chip, Flint produced about 131 words or code segments per second, while in tests with an Nvidia RTX Spark card it exceeded 150 such elements per second.
READ ALSO
General Motors unveils new platform and more powerful mainframe
Funding and background
Recursive, founded last year and led by Richard Socher, a former chief scientist at Salesforce, has raised $650 million in funding at a $4,65 billion valuation, backed by investors including GV, Nvidia and AMD Ventures. It also signed a $410 million computing power deal with Amazon Web Services in June.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
Top comments (0)