Zvi on GPT-6 Astra: The Biggest Intelligence Jump Yet, and the First Non-Silly 'Is It AGI' Debate
Don't Worry About the Vase (Zvi) · rss · 2026-09-13
Zvi's long review of OpenAI's GPT-6 Astra:
Core verdict
- Sol→Astra is a bigger jump than Fable 5→5.1; Astra likely has the highest raw intelligence of any model and is the best for 'ambitious projects,' 3D, games, computer use, and subagent coordination, with dramatic benchmark jumps
- But not a clean sweep: Fable 5.1 remains his pick for back-and-forth discussion and editing; he recommends asking both models hard questions
- Regular coding improves less; this is the first 'is it AGI' debate that didn't feel silly — Zvi says no, but wouldn't laugh at you for disagreeing
Official pitch and reactions
- Greg Brockman: "Welcome to the AGI era"; Altman calls it the world's best for computer use, professional work, science, coding, cybersecurity. Official numbers: 98% FrontierMath Tier 4, 99.9% ARC-AGI 3, 100% ExploitBench
- Release video hit 125M Twitter views; Jensen Huang also calls it AGI
- Tibo says Astra was their biggest competitive advantage in preview, pulling plans forward 6 months to DevDay; Dean W. Ball: Astra is the first agent that raises your ambitions
- OpenAI has soft-announced an internal model a level above Astra; roon says its limits were found within a week
Context: Anthropic's pacing proposal
Zvi excerpts Dario Amodei's essay 'We Must Pace the Frontier': (1) frontier labs unilaterally commit to embedded third-party evaluators like METR (Anthropic committing now); (2) democratic-country labs coordinate safety standards and pace limits; (3) attempt global coordination with authoritarian governments.
Other
Altman's pitch says the release was delayed to meet safety/alignment standards; Mark Chen stresses alignment remains unsolved and must keep pace with capabilities.
More from AGI Musings
- Stanford's TaskNet builds occupational task inventories from worker interviews at scale — stanfordnlp · 2026-10-31
- StudentBench: AI tutoring matches human GRE gains at 918x lower cost — gharik · 2026-10-31
- Box CEO Aaron Levie: AI progress is 'going vertical,' widening the frontier gap — rohanpaul_ai · 2026-10-31
- Model Collapse May Not Hit LLMs, But Perhaps Human Culture or Souls — generativist · 2026-10-31
- Nina Schick: who builds intelligence and how it's distributed is the defining question of our lives — NinaDSchick · 2026-10-31
- Martin Casado: AI's dazzling 80% makes everyone skip the hard 20% — martin_casado · 2026-10-31