Recursive launches Flint, a laptop coding model it says was largely built by AI
- ID
- 38042
- Status
- summarized
- Published
- 22 Oct 2026, 4:27 AM
- Fetched
- 22 Oct 2026, 5:17 AM
- Provider
- TechCrunch
- Category
- technology
- Original URL
- https://techcrunch.com/2026/10/21/recursive-launches-flint-a-laptop-coding-model-it-says-was-largely-built-by-ai/
- Source URL
- https://techcrunch.com/feed/
Summary
- Score
- 6.5
- Created
- 22 Oct 2026, 5:17 AM
- Tags
- Audience
- developersvibe_codersai_ml_learnerssaas_founders
What happened
Recursive released Flint, a free Apache 2.0 coding model on Hugging Face that runs on a laptop and uses a mixture-of-experts design with 40B total parameters but only 2.5B active per task. The company says its Eureka Machine automated research system helped build Flint from scratch in under four months, and that the model can run offline to keep code private and avoid per-token cloud fees. Recursive emerged from stealth in May with $650M at a $4.65B valuation, led by CEO Richard Socher.
Why it matters
You can download and test Flint locally without per-token fees, but the practical question is whether a 2.5B-active coding model is good enough for your day-to-day tasks; benchmark it on your own repo before switching any workflow, and treat the 'largely built by AI' claim as a vendor claim, not an independent result.
Discussion angle
What would make a 2.5B-active laptop coding model good enough for your daily work, and which tasks would you still keep on cloud models?