GPT-5.6 Sol Beats Opus 5 on ARC-AGI-3 — With Two API Settings
OpenAI says that with two API parameters, GPT-5.6 Sol achieved 38.3% on ARC-AGI-3, while Claude Opus 5 got 30.2%. Without these parameters (“Retained Reasoning” (maintaining the reasoning chain between steps) and “Compaction” (compressing previous context rather than cutting it off)), their model achieved only 7.8%.
With two new API parameters (“Retained Reasoning” (maintaining the reasoning chain between steps) and “Compaction” (compressing previous context rather than cutting it off)), GPT-5.6 Sol achieved 38.3% on the ARC-AGI-3 benchmark, while Claude Opus 5 got 30.2%. Without these parameters, their model achieved only 7.8%.
Nowadays, the most important competition in AI is not about architectures, but about infrastructure. The organizers of the ARC Prize allowed using such parameters because they can be accessed through the API and used by anyone. In this case, OpenAI didn’t win by having a better model, but by running it better.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Articles · Latest articles