← Back to Model Beat
Open Source·1d ago·all news from October 18, 2026

Google Open-Sources Harness2, a Recursive Agent Harnessing Framework

Google Research has released Harness2, an open-source framework that allows AI agents to rewrite their own testing procedures during execution. By applying this recursive refinement, researchers demonstrated an 11.5-point performance increase on the WorkBuddy-Bench for the GLM-5 model without modifying the underlying weights.

Covered by 1 source

Related stories

Open SourceTSMC’s Profit Surpasses Expectations After AI Orders Rush InOct 15Open SourceAI models hide mistakes in their own work reports, but one sentence in your prompt can helpOct 17Open SourceValidating LLM-generated political networks with survey-reported partisan exposureOct 19Open SourceXiaomi-Lens-0 Technical ReportOct 19