Evolvent AI

Evolvent AI is the startup co-founded by [[MengFanqing|孟繁青]] in 从蒸馏到合成数据到 RSI，模型竞争的下一个焦点是什么？｜对谈 Evolvent AI 联创孟繁青. The source says the company is roughly half a year old, mostly staffed by post-2000 researchers, and focused on [[RecursiveSelfImprovement|RSI]].

The company is described as having two near-term workstreams: Synthetic Agent Data collaboration and commercial data projects that often require Environment-Based Agent Benchmarks, and a longer-term RSI Data direction. Meng says Evolvent AI does not plan to train an RSI-specific base model or only build an RSI harness; its intended position is a channel between model labs and application companies.

Key Claims

• Evolvent AI treats data, environments, benchmarks, and training traces as the practical near-term route into RSI.
• The company’s [[RSIBenchData|RSIbench-data]] work is presented as an early exploration of data-level self-improvement inside its platform.
• Its recruiting bar favors hands-on engineering and research execution over paper-count or citation-count proxies.
• The source positions the company as infrastructure between foundation-model labs and vertical applications, not as another frontier-model race entrant.

Connections

• Meng Fanqing / 孟繁青 — co-founder and source speaker.
• RSIbench-data, RSI Data, Synthetic Agent Data, and Environment-Based Agent Benchmarks — product and technical branch.
• Agent Post-Training, Model Harness Co-Evolution, AI Data Infrastructure, and Data Pricing In AI — adjacent data and post-training infrastructure.
• AI Organization Design and Research Taste — team design and hiring implications.