# Recursive AI Built Flint Coding Model

> Recursive's AI created most of its new Flint coding model without human input.
- Title: Recursive AI Built Flint Coding Model · Startup Fortune
- Summary: Recursive's AI created most of its new Flint coding model without human input. Recursive's AI system developed Flint, an open-weight coding model that runs…
- Keywords: recursive, flint, eureka machine, coding model, ai self-improvement, open-weight model, swe-bench, terminal-bench, startups, business, Built, Coding
- Source: Startup Fortune — https://startupfortune.com/recursive-says-ai-built-most-of-its-new-flint-coding-model
- Published: 2026-10-21T19:42:48+00:00
- Read time: 1 min
- Topics: recursive, flint, eureka machine, coding model, ai self-improvement, open-weight model, swe-bench, terminal-bench, startups, business
## Recursive's AI Built Flint
Recursive's AI system developed Flint, an open-weight coding model that runs locally on laptops and outperforms benchmarks.
## Flint's Development Process
Nearly every stage of Flint's pipeline, training method, and model architecture came from ideas the Eureka Machine generated and tested itself.
## Flint's Technical Specs
Flint is a 40-billion-parameter mixture-of-experts model with 2.5 billion active parameters, running on a laptop with an M4 Pro chip.
## Performance Metrics
Flint hits over 125 tokens per second on a MacBook Pro with an M4 Pro chip and more than 150 on Nvidia's RTX Spark.
## Benchmark Results
Recursive claims Flint leads its size class on SWE-Bench Pro, SWE-Bench Verified, and Terminal-Bench tests for coding tasks.
## Independent Verification
Flint's benchmark results were generated by Recursive itself, not an independent lab, and have not been verified by outsiders yet.
## Recursive's Self-Improvement Claim
Recursive says its AI system can find weaknesses, write benchmarks, and rewrite code to improve without much human input.
## Key takeaway

Recursive's AI system developed Flint, an open-weight coding model that runs locally on laptops and outperforms benchmarks.