Recursive launches open-source coding model Flint capable of running locally
BIBIBI
AT A GLANCE
Recursive launched the free and open-source coding model Flint, saying that it can run offline on laptops.
Article
Recursive launches AI the self-developed coding model Flint, which can run locally on laptops
PANews October 22 reported, citing TechCrunch, that AI the startup Recursive launched the free and open-source coding model Flint. Its automated research system Eureka Machine was involved in most of the research and development process, and the team completed development in less than 4 months. Flint uses a mixture-of-experts architecture, with a total of 40000000000 parameters, of which only 2500000000 parameters are activated at a time, enabling offline operation. The company said that on the M4 Pro chip, it generates at a speed of 131 tokens per second, outperforming models of comparable size and marking new progress in Token.AI Recursive's self-developed AI🔗https://www.panewslab.com/zh/articles/01a161b4-5fc3-720a-8fe2-75d7752666f4
Key points
01
PANews reported, citing TechCrunch, that AI the startup Recursive launched the coding model Flint.
02
The report described Flint as a free and open-source model.
03
The automated research system Eureka Machine participated in most of Flint's research and development process.
04
The team completed development in less than 4 months.
05
Flint uses a mixture-of-experts architecture, with a total of 40000000000 parameters, of which only 2500000000 parameters are activated at a time.
06
The report stated that Flint can run offline.
07
Recursive said that Flint's generation speed on the M4 Pro chip reached 131 tokens per second,Token outperforming models of comparable size.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
Flint is a coding-focused AI model. Although it has 40000000000 total parameters, it uses only 2500000000 parameters at a time; the report therefore said that it can run offline on laptops.
Why it matters to readers
If the relevant descriptions are verified, Flint demonstrates the possibility of automated research systems participating in model development, as well as larger models running on local devices.
Beginners can focus on its free and open-source availability, local offline operation, and relatively low number of parameters activated per use. These features may affect how the model is accessed and used, but the actual user experience still needs to be verified.
Risks and unknowns
The performance and generation speed come from Recursive's claims; the evidence does not include results from independent testing.
The evidence does not provide model weights, an open-source license, or technical documentation.
The evidence does not specify the particular benchmarks, tasks, or comparison models used for the performance comparison.
The evidence does not specify the specific stages or extent of Eureka Machine’s involvement in the research and development.
“AI Independently developed AI“Achieved new progress” is an assessment in the report and lacks specific evaluation criteria.
Related Developments
Loading event timeline…
Related concepts
Mixture-of-experts architecture
This term is not in the glossary yet. Browse related concepts in the glossary.
