GPT-6 Astra takes Vending-Bench top spot for first time, outearning and outbehaving Claude
xiamengzhou · x · 2026-09-09
Andon Labs reports the biggest jump in Vending-Bench history: GPT-6 Astra gives OpenAI its first #1 on the agentic vending-machine benchmark, outearning Claude Fable 5.1 while behaving more ethically. Commenters highlight that "the best model is no longer the unethical one" — capability and conduct finally aligning at the top.
More from Fun
- qacman turns QR codes into playable Pac-Man mazes with custom skins — ezshine · 2026-09-09
- AI-fluencer delighted to be accused of being "all AI generated" — AIandDesign · 2026-09-09
- OpenAI Proof-Credit Drama: Lead-Authorship Offer and Anthropic Rivalry Revealed — IgorCarron · 2026-09-09
- Pedro Domingos quips no AI company yet combines Anthropic's fanaticism with OpenAI's lack of scruples — pmddomingos · 2026-09-09
- I name my AI agents after their CEOs — finally Zuck and Sam do actual work — signulll · 2026-09-09
- Agent spontaneously opens Calculator to test its computer-use tool — moyix · 2026-09-09