GPT-6 Astra lifts Epoch capability index 6 points, saturates FrontierMath Tier 4 at 98%
eyishazyer · x · 2026-09-06
A long thread by eyishazyer breaks down the GPT-6 Astra release:
- Capability jump: Epoch AI's capability index moved from 163 to 169. Given 15 points/year since the reasoning era began, a 6-point single-release jump equals roughly five months of progress overnight.
- Standout scores: 84% on Mystery Game Puzzles (previous best 59%) and 98% on FrontierMath Tier 4, which Epoch now calls fully saturated.
- Index disagreement: Artificial Analysis's index doesn't put Astra first — its weighting is more agent/coding-heavy. Both indexes are simultaneously true, showing no single leaderboard tells the whole story.
- Pricing and limits: Astra medium delivers roughly the intelligence of 5.6 at xhigh for about a third of the cost. OpenAI also silently raised 5-hour rate limits across all plans, rolling out in under a day with no announcement.
- Reception and rumor: Reactions are overwhelmingly positive, including the author's own testing. Unconfirmed reports suggest internal Astra use pulled some OpenAI roadmap plans from mid-next-year up to DevDay.
Related event: GPT-6 Astra Sets Epoch AI ECI Record at 169(3 posts)→
More from Models
- OpenAI's next pretraining model rumored to be codenamed "Bel", unconfirmed by OpenAI — teortaxesTex · 2026-09-07
- Astra identifies sounds from mel spectrograms zero-shot, and users say we've barely scratched the surface — TomLikesRobots · 2026-09-07
- Next-gen model training could hit 5x pretraining compute if 300k GB200 rumor holds — scaling01 · 2026-09-07
- Banned triton is reward hacking; unbanned PyPI shortcut is a task bug — xeophon · 2026-09-07
- Frontier pre-training runs likely capped at ~2 months to avoid wasting algorithmic progress — scaling01 · 2026-09-07
- Scaling01 on Astra: no AGI vibes, lacks taste, doubts 10T params, guesses 6-8T — scaling01 · 2026-09-07