Harvey partnered with baseten on M&A diligence agents on 2026-09-08

Harvey said on 2026-09-08 that it partnered with baseten to post-train recursive language model agents for M&A diligence. The company said it found that model-harness co-optimization meaningfully improves agent performance in long-horizon environments and brings it closer to agents that can complete M&A diligence.

Gabe Pereyra said the work with baseten and baselabs produced a recursive language model harness. According to Pereyra, the harness lets a root agent delegate document review to sub-agents and combine their findings into a diligence memo. His post identified the work as Post-training Qwen 3.5.

Harvey’s account described the partnership as focused on post-training recursive language model agents, while Pereyra described the harness and its document-review process. Both posts attributed the work to Harvey, baseten and baselabs.