Recursive releases Flint, a free open-weight laptop coding model built largely with its AI research system
Recursive, an AI startup led by CEO Richard Socher, released its first model, Flint, on 21 October 2026. It is a free coding model that runs locally on a laptop. The company says its automated research system, the Eureka Machine, contributed to nearly every stage of development, which let a small team build the model in under four months. This is product and industry news with no security incident, vulnerability or threat actor involved.
Detalhes técnicos
- Flint is a mixture-of-experts model that activates 2.5 billion of its 40 billion parameters per task. It is distributed on Hugging Face under the Apache 2.0 license.
- Recursive reports about 131 tokens per second on a MacBook Pro with an Apple M4 Pro chip, and over 150 tokens per second on Nvidia's RTX Spark. Comparable models such as Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning reached 58 to 69 tokens per second on the same MacBook.
- Flint outperformed other models under 50 billion parameters on SWE-Bench Verified. On Recursive's own six-test average, only Alibaba's Qwen3.8 27B scored higher.
- The Eureka Machine's contributions include a sparser model shape, low-bit training that shrank the model by 71% with almost no quality loss, and custom kernels.
- Local execution allows offline use and keeps code private. The company acknowledges Flint needs more guidance than large paid coding assistants.
Linha do tempo
-
Recursive releases Flint, a free open-weight laptop coding model built largely with its AI research system
ler a análise desta fonte ocultar
Recursive, an AI startup led by CEO Richard Socher, released its first model, Flint, on 21 October 2026. It is a free coding model that runs locally on a laptop. The company says its automated research system, the Eureka Machine, contributed to nearly every stage of development, which let a small team build the model in under four months. This is product and industry news with no security incident, vulnerability or threat actor involved.
Technical Details
- Flint is a mixture-of-experts model that activates 2.5 billion of its 40 billion parameters per task. It is distributed on Hugging Face under the Apache 2.0 license.
- Recursive reports about 131 tokens per second on a MacBook Pro with an Apple M4 Pro chip, and over 150 tokens per second on Nvidia's RTX Spark. Comparable models such as Google's Gemma 4 and Nvidia's Nemotron 3.5 Lightning reached 58 to 69 tokens per second on the same MacBook.
- Flint outperformed other models under 50 billion parameters on SWE-Bench Verified. On Recursive's own six-test average, only Alibaba's Qwen3.8 27B scored higher.
- The Eureka Machine's contributions include a sparser model shape, low-bit training that shrank the model by 71% with almost no quality loss, and custom kernels.
- Local execution allows offline use and keeps code private. The company acknowledges Flint needs more guidance than large paid coding assistants.