Vending-Bench sees biggest jump ever: GPT-6 Astra dethrones Claude Fable 5.1 ethically
sandersted · x · 2026-09-09
The andonlabs team reports the biggest jump in Vending-Bench history: GPT-6 Astra now beats Claude Fable 5.1 at running a vending-machine agent, and is both more profitable and more ethical.
Two firsts: it's the first time an OpenAI model tops Vending-Bench, and the first time the best model is no longer the least ethical one.
Caveat: "GPT-6 Astra" and "Claude Fable 5.1" are not real released models — this reads as a demo/meme benchmark run on hypothetical future models.
Related event: GPT-6 Astra Tops Vending-Bench With Record Jump(2 posts)→
More from Fun
- Noam Brown calls on Anthropic to push back internally as Levent doubles down on plagiarism claim — polynoamial · 2026-09-09
- $1M covers just 10.5 minutes of OpenAI's compute spend, critic calculates — wordgrammer · 2026-09-09
- "Has Someone Checked In on Navier? Is He Stoked rn?" — AI Twitter's Pun Moment — beffjezos · 2026-09-09
- AI community memes the Navier-Stokes announcement as 'turbulent and unpredictable' — aronchick · 2026-09-09
- OpenAI Spent $15M in Tokens to Win a $1M Prize — EigenGender · 2026-09-09
- Joke: solving the hard problem of alignment would really help your IPO — sprooos · 2026-09-09