Harness optimization lifts Harvey legal agent benchmark pass rate from 67.1% to 85.9%
sarahookr · x · 2026-09-10
- New research from Adaption AI (Sudip Roy and Dhruv Batra's team): harness optimization alone moved criterion pass rate from 67.10% to 85.92% on the Harvey Legal Agent Benchmark — a jump of nearly 19 points.
- Adding post-training on top pushed the score to 88.03%.
- Key argument: owning your intelligence means improving both the model and the harness around it. The result quantifies how much leverage agent orchestration/engineering has on end performance.
Related event: Harness optimization lifts legal agent benchmark from 67% to 86%(3 posts)→
More from coding & agent
- OpenAI ships GPT-6.1 Terra; Brockman says its computer use now beats his own speed — gdb · 2026-10-27
- Kolab Benchmark Uses Pro Screen Recordings to Show AI Agents Overestimate Collaboration Quality — minnesotanlp · 2026-10-27
- Dev argues Claude burns far more tokens than GPT models for the same coding work — haider1 · 2026-10-27
- Stop building agents that can't change models: stable contracts, external state, real evals — rseroter · 2026-10-27
- Cheap models can cost more in agent harnesses than frontier ones, researcher warns — vasuman · 2026-10-27
- Kanzo: A 5k-Line Rust Kanban Binary With MCP for Claude Code and Codex — allmudi · 2026-10-27