Harness-1: A 20B Stateful Retrieval Subagent Trained with Reinforcement Learning

Harness-1：基于强化学习训练的有状态搜索20B检索子智能体

UIUC and Chroma released Harness-1, a 20B-parameter retrieval subagent trained with reinforcement learning inside a stateful search harness, reporting 0.730 average curated recall across 8 benchmarks, 11.4 percentage points above the next-best open-source subagent and behind only Opus-4.6.

Why it matters: HKR-H/K/R all pass: Harness-1 has a clear RL retrieval-agent mechanism and benchmark numbers. It stays in 78–84 because this is a subagent research/open-source release, not a major lab model launch.