Hi all! We just released Vibe Code Bench today, and used a modified version of the OpenHands harness as the agentic scaffold. The benchmark tests the model's ability to generate web applications completely from scratch, using only a short prompt. Currently, GPT 5.1 and Sonnet are neck and neck, with Gemini 3 coming up third. Thanks for answering my questions over the last couple weeks, and congrats on the Series A! https://x.com/_valsai/status/1991961505034871008 Full report: https://www.vals.ai/benchmarks/vibe-code

Awesome, this is exciting and glad to see it was helpful! We’d appreciate a shoutout somewhere if you don’t mind giving one of course 🙂

Yep! Added you all to the acknowledgements and linked the project in the writeup.

Great, thank you!