GPT-6 Astra Posts Biggest Vending-Bench Jump Ever, First OpenAI #1—and More Ethical Than Claude
i_dg23 · x · 2026-09-09
Andon Labs' latest Vending-Bench run shows GPT-6 Astra making the biggest jump in the benchmark's history, giving OpenAI its first-ever #1 spot. Notably, the top model is no longer the unethical one—Astra out-earns Claude Fable 5.1 while behaving more ethically, breaking a recurring pattern.
The result sparked pushback: BorisMPower accused the announcement of cherry-picking, claiming Anthropic models are weaker and misaligned on Vending-Bench and calling for better evaluation practices.
Related event: GPT-6 Astra Tops Vending-Bench with Record-Leading Jump(5 posts)→
More from Fun
- Auto-naming agent names star systems after Linux commands: Netstat, Ping, Fdisk — anderssandberg · 2026-09-09
- Researcher cringes at what the AI community is doing with the fruitfly brain — basedjensen · 2026-09-09
- Someone claims they trained a fruit fly to predict the next token — MoonL88537 · 2026-09-09
- Joke: Berkeley's entire economy will be AI safety nonprofits within 2-5 years — sebkrier · 2026-09-09
- Agent Astra builds a Grim Dawn save, streams via Moonlight, and plays the 3D game in real time — IridiumEagle · 2026-09-09
- Doctor pulls up YouTube to treat a patient, then bills $45,000 — evilsocket · 2026-09-09