OpenAI Quietly Altered GPT-6 Astra Benchmark Scores: Fortune

Fortune reported that OpenAI quietly edited multiple benchmark figures around GPT-6 Astra's launch, in some cases improving Astra's numbers while downgrading competitor Anthropic's. The hallucination rate reportedly changed from 4.2% to 2% at one point, raising transparency concerns.

2026-09-05 ~ 2026-09-05 · 3 related posts

Full story(7 episodes)→

1 near-duplicate retellings: mark_k