ARC Prize publishes GPT-6 Astra results on ARC-AGI-3 benchmark

vignesh_warar · hn · 2026-09-04

The ARC Prize team has published a blog post evaluating OpenAI's new GPT-6 Astra model on ARC-AGI-3, the benchmark focused on abstract reasoning and agentic capability. See arcprize.org/blog/astra for the full results and analysis.

Original post →

More from Models

Models channel →