It's AGI at 99.9%. Unplug their private harness and it's 62.7%.

OpenAI President Greg Brockman says GPT-6 Astra marks the beginning of "the AGI era." On the ARC-AGI-3 benchmark it scores 99.9% - but only while using OpenAI's own private software harness; on the b…

Source: Live Science, 'OpenAI claims we've entered the AGI era - has GPT-6 Astra really demonstrated general intelligence?' (Adam Shepherd, 2026-10-08). Verified facts: ARC-AGI-3 headline score of 99.9% was achieved using OpenAI's proprietary 'Provider Adapter' software harness; on the benchmark's neutral 'Standard' harness the same model drops to 62.7%. An embargoed pre-launch draft listed the ARC-AGI-3 score at 98.6%; it became 99.9% on the live site. Stanford doctoral researchers Anka Reuel and Mike Hardy flagged OpenAI quietly revising metrics post-launch, including halving Astra's reported hallucination rate from 4.2% to 2.0% and then reverting it; they call the practice 'benchmaxxing.' ARC Prize president Greg Kamradt said saturating the benchmark 'would not represent proof of achieving AGI' and that Astra is 'meaningful progress' but 'we are not claiming that it is AGI.' OpenAI touts real skills: 3D CAD code, PCB layout, 3D-model-to-game environments, prompt-to-web-app deploys. Brockman at the 2026-09-03 launch: 'when was it really that AGI was created ... it might be about this model.'

More memes on lilmeme.fun

Open on lilmeme.fun