DeepMind Paper Evolves Agent Harness Together with Weights

Google DeepMind's new paper Evo² shows that co-evolving an agent's harness with model weights matters as much as the model itself, boosting SWE-bench scores from 24.4% to 48.3% on GPT-5-mini.

2026-10-13 ~ 2026-10-13 · 2 related posts

• DeepMind paper co-evolves agent harness and weights, lifting SWE-bench 41.0% to 48.3% — omarsar0 · 2026-10-13
• Google paper: evolving the agent harness doubles GPT-5-mini's SWE-bench score to 48.3% — bittingthembits · 2026-10-13