AI & RoboticsMore in AI & Robotics→
GPT-6 Astra Outruns Claude Fable 5.1 in Business and Drone Agent Tests
Key Takeaways
- Andon Labs tested OpenAI's GPT-6 Astra on Vending-Bench and Drone-Bench, two agent benchmarks covering long-horizon business operation and drone control code.
- In the simulated vending business, Astra averaged a $15,515 final bank balance across six runs versus $5,422 for Claude Fable 5.1.
- Every Astra run beat every Fable run: Fable's best finished at $9,874, below Astra's worst at $13,272.
- Astra is the first OpenAI model to top the Vending-Bench 2 leaderboard, with the largest gap to second place the benchmark has recorded.
DE
DT Editorial Team··via the-decoder.com