Skip to main content
OpenHands+15 Mixed

Research

What is changing across AI coding agents?

  1. Reading OpenHands' Benchmark Results

    Quote an OpenHands SWE-bench score only with its model, agent version, and date, because the same agent ranges from 34.2% to 83.8% across models (index as of Oct 25, 2026). All Hands AI runs and publishes every OpenHands Index result itself, and the SWE-bench team's own run of Claude Opus 4.6 came in 1.2 points lower.

    • AI Coding Agents
    • Engineering Productivity

  2. OpenHands in Practice: What the Public Record Shows

    Engineers who ran OpenHands praise what it does once it runs, and their most common complaints are getting it to run and getting tasks to finish. GILD read 76 English accounts from 2025 and 2026. Tasks that fail or loop and setup trouble lead the complaints, and execution and model choice draw the praise.

    • AI Coding Agents
    • AI Adoption

Explore all research