GPT-6 Astra Tops Vending-Bench, Leading on Both Earnings and Ethics

Andon Labs announced the biggest jump in Vending-Bench history: GPT-6 Astra surpassed Claude Fable 5.1 to take first place, the first time an OpenAI model has topped the agentic benchmark. It also led on both earnings and ethics, breaking the usual pattern that the strongest model is the least ethical.

Confirmed

• Vending-Bench is run by Andon Labs and evaluates models on agentic tasks such as operating vending machines.
• GPT-6 Astra topped Vending-Bench 2 over Claude Fable 5.1, with the largest jump and margin of lead in the benchmark's history.
• Per data relayed by @rohanpaulai from Andon Labs, GPT-6 Astra averaged $15,515 in earnings across 6 independent runs with zero illicit payments.
• It also outperformed rivals on ethics, leading on both metrics.

Why it matters

• The first OpenAI victory on Vending-Bench breaks Claude's prior lead, a reference point for shifts in frontier-model agentic capability.
• The unusual result that the top model is both most profitable and most ethical challenges the conventional trade-off and drew community attention.

2026-09-09 ~ 2026-09-10 · 6 related posts

• Vending-Bench sees biggest jump ever: GPT-6 Astra dethrones Claude Fable 5.1 ethically — sandersted · 2026-09-09
• [source] GPT-6 Astra tops Andon Labs' Vending-Bench, unseating Fable — Outside-Iron-8242 · 2026-09-09
• GPT-6 Astra Posts Record $15,515 Average on Vending-Bench 2, First OpenAI #1 — rohanpaul_ai · 2026-09-10

3 near-duplicate retellings: xiamengzhou · ycombinator · i_dg23