Harvey post-trains Qwen3.5 orchestrator: rubric pass rate jumps from 29.9% to 63%, beats Claude Code
SergioPaniego · x · 2026-09-09
Harvey published a blog post on post-training open-model agents for end-to-end M&A diligence. They trained a Qwen3.5-122B-A10B orchestrator, raising the rubric criteria pass rate from 29.9% to 63.0% on 50 held-out LAB Diligence data rooms — outperforming Claude Code and Codex, they claim.
Harvey's Tenet research preview emphasized model-harness co-optimization as key to complex legal tasks; this post is that approach applied in a real legal workflow. merve (Hugging Face) reshared it noting that post-training open-model agents is where the moat lies.
More from coding & agent
- Stanford's PaperFormalizer turns research papers into Lean-verified versions with author in the loop — StanfordUniversity · 2026-10-27
- Hugging Face TRL makes score centering + masked importance sampling default in async GRPO — _lewtun · 2026-10-27
- AutoResearch agent's weird training curve gets dug into instead of discarded — ZimingLiu11 · 2026-10-27
- Firecrawl offers 2.5x credits to a16z Speedrun founders, unveils $75M Series B — devdigest · 2026-10-27
- Chums: open-source Apple Watch app puts Claude Code and Codex on your wrist — brandon_galang · 2026-10-27
- Open-source Clairic Studio turns pasted text into typed data and working code in 5 languages — dfinke · 2026-10-27