Press Space to continue
GPT-6 Astra Uses Reasoning Tokens Far Less Than GPT-5.6 Sol
Press Space to continue
With low reasoning effort in our standard harness, GPT-5.6 Sol used reasoning tokens in 98% of the actions it took while playing ARC-AGI-3 public games. With the same low reasoning effort, GPT-6 Astra used reasoning tokens in only 27% of its actions (!). We saw a similar pattern with the semi-private games, suggesting this difference was not driven by public benchmark contamination and may instead reflect architectural changes in how the GPT models work.
mikeknoop@mikeknoopSep 4
New chart from Astra testing on ARC v3 Astra often emits zero reasoning tokens per action at lower reasoning levels. We've never seen this before. And surprisingly Astra low is 2X more accurate than Sol max. This suggests Astra is leveraging a secondary test-time adaptation scaling axis, presumably latent space reasoning.