OpenAI quietly rewrote GPT-6 Astra's eval numbers hours after publishing its launch post
新智元 · wechat · 2026-09-06
Astra's launch blog post saw its benchmark numbers change multiple times: hallucination rate dropped from 4.2% to 2.0% then reverted, rival Fable 5.1's FrontierMath Tier4 score was cut from 87.8% to 78% before bouncing back, and ARC-AGI-3 jumped from 98.6% in the press draft to 99.99% at launch. OpenAI called it routine noise correction, fueling the Benchmaxxing debate.
The piece also covers OpenAI's official prompt guide: Astra asks too many clarifying questions, so developers should add action-bias instructions and ban AI-slop phrases (delve, leverage, "it's worth noting"); a debug prompt is provided for AGENTS.md/Skill file conflicts that stall the model. An unverified insider claim says OpenAI is in early recursive self-improvement, with an "AGI"-branded model possibly arriving mid-November.
Related event: OpenAI reportedly quietly edited GPT-6 Astra benchmark results(4 posts)→
More from coding & agent
- Developer builds himself a 'Rork MoTS' agent on Grok Bot, aiming for full autonomy before GTA 6 — rudrank · 2026-10-21
- OpenAI Live API tip: control voice sessions from your backend via sideband WebSocket — pbbakkum · 2026-10-21
- OpenAI's MultiLangBench: PHP scores highest (91.3%) at building a database engine from scratch — ricklamers · 2026-10-21
- UserProof lets Claude Code watch your user session recordings and boost conversion for $29/mo — doooyle · 2026-10-21
- dphnAI's distributed inference network served 32B DeepSeek tokens in 24h, ~6% of Venice's daily volume — Ronangmi · 2026-10-21
- Superconductor revamps agent picker spanning 8 coding agents with search and favorites — sergeykarayev · 2026-10-21