What's Changed

This release contains improvements that allow OpenHands to score 53% on SWE-Bench Verified & 41.7% on SWE-Bench Lite, making it the best OSS AI SWE system now: https://github.com/swe-bench/experiments/pull/97

Added

• Add function calling to CodeAct 2.1 by @xingyaoww in https://github.com/All-Hands-AI/OpenHands/pull/4537
• Display error messages in the chat by @amanape in https://github.com/All-Hands-AI/OpenHands/pull/4509
• Support prompt caching on the latest claude-3-5-sonnet-20241022 by @xingyaoww in https://github.com/All-Hands-AI/OpenHands/pull/4552

Changed

• Pared down the backend logging to make it less verbose and more informative by @rbren in https://github.com/All-Hands-AI/OpenHands/pull/4563

Fixed

• Split bash commands by the new line character by @tofarr in https://github.com/All-Hands-AI/OpenHands/pull/4462
• Bugfix for loading the /app page directly by @rbren in https://github.com/All-Hands-AI/OpenHands/pull/4547
• Fix for connection leak by @tofarr in https://github.com/All-Hands-AI/OpenHands/pull/4560
• Fix for crash on some models on OpenRouter by @enyst in https://github.com/All-Hands-AI/OpenHands/pull/4641
• Fix compatibility for models like llama on Groq that don't support vision or tools by @enyst in https://github.com/All-Hands-AI/OpenHands/pull/4644

New Contributors

• @msehsah1 made their first contribution in https://github.com/All-Hands-AI/OpenHands/pull/4580

Full Changelog: https://github.com/All-Hands-AI/OpenHands/compare/0.11.0...0.12.0