Flint
Recursive released Flint, its first model, a 40B mixture-of-experts open-weight coding model that runs locally at over 125 tokens per second on a MacBook Pro with an M4 Pro chip.
Flint is the first public model from Recursive, a startup focused on recursive self-improvement. A small team built it in under four months, using an automated research system the company calls the Eureka Machine. Per an LM Studio post quoted on Techmeme, it is a 40B MoE with 2.5B active parameters, a 262K token context window and an Apache 2.0 license, aimed at local agentic coding.
- Date
- Wednesday 21 October 2026
- Lab
- Recursive
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Generation speed on MacBook Pro (M4 Pro) | over 125 tokens per second company claim | company |
| Parameters | 40B total, 2.5B active active count from an LM Studio post quoted on Techmeme | company |
| Context window | 262K tokens from an LM Studio post quoted on Techmeme | company |
| Speed vs comparable models | up to 2x company claim, not independently tested | company |
Memory needs are reported inconsistently, 48GB by Crypto Briefing and about 24GB by LM Studio. Weights are on Hugging Face. Recursive has reportedly raised $650M at a $4.65B valuation.
Sources
This record was partly confirmed: some claims could not be checked on 21 October 2026. How we check