Researchers train open source AI search agent Harness-1, outperforming GPT-5.4 on relevant information recall
Summary
A joint research collaboration between UIUC, UC Berkeley, and Chroma released Harness-1, a 20-billion parameter open-source search agent built on OpenAI's gpt-oss-20B model. Harness-1 achieves 73% average on recalling relevant information from a curated dataset, outperforming GPT-5.4 (70.9%) and Tongyi DeepResearch 30B by 11.4 percentage points. The model and code are available under Apache 2.0 on Hugging Face. It also validates the Tinker distributed AI training API by Thinking Machines.
Why it matters
This event demonstrates open-source AI models surpassing closed-source frontier models on specific retrieval tasks and validates distributed training infrastructure.
Source links
Market reaction
The following is market reference information for related companies, and does not constitute investment advice.