Open-source Gear framework: a meta-learning tool for RSI

Developers built Gear, an open-source meta-learning framework aiming to be a scikit-learn for RSI. In benchmarks, Gear-optimized DSH improved pass rates from 27.0% to 40.0% on AutomationBench Marketing and from 52.8% to 62.9% on Terminal-Bench 2.1.

• Surge AI Launches DAYJOB: Tax Agent Benchmark2026-10-28
• Google open sources Harness2 for self-improving AI agents2026-10-19
• Open-weight models edge past Sonnet 5 on some benchmarks2026-10-16
• LegalOn cuts daily Codex costs by 65%2026-10-09
• SWE-Race benchmark tests coding agents on real concurrency bugs2026-10-06