Leaked GPT-6 Astra benchmarks show ARC-AGI-3 jumping from 7.8% to 98.6%
burhop · x · 2026-09-04
Screenshots purportedly from OpenAI's website show GPT-6 Astra benchmarks (unverified):
- ARC-AGI-3: jumps from 7.8% to 98.6%, claimed near-saturation
- ExploitBench: 100% score
- Rollout: limited organizations today, then all ChatGPT Plus/Pro/Business/Enterprise users in the coming days, plus OpenAI API and AWS
The poster calls it an insane jump and declares "AGI is here." Note this is third-party reposting, pending official confirmation from OpenAI.
Related event: GPT-6 Astra Launches with Benchmark Leaks: ARC-AGI-3 Hits 98.6%(33 posts)→
More from Models
- GPT-6 Astra shows substantially lower chain-of-thought monitorability — scaling01 · 2026-09-04
- johnlindquist shares what he built with early-access GPT-6 — johnlindquist · 2026-09-04
- After GPT-5.6 wiped his Mac, early reviewer says Astra brought him back to OpenAI — mattshumer_ · 2026-09-04
- Allie K. Miller's early Astra review: stunning computer use, still fails wordplay — alliekmiller · 2026-09-04
- Matthew Berman demos Astra-built 3D liminal horror game and mini biome world — MatthewBerman · 2026-09-04
- Matthew Berman has Astra generate a 3D ASCII art city — MatthewBerman · 2026-09-04