How enabling two settings tripled our scores on the ARC-AGI-3 benchmark

TL;DR

Enabling two specific API settings significantly boosted GPT-5.6's performance on the ARC-AGI-3 benchmark, tripling scores while improving efficiency by enhancing reasoning capabilities and compacting data retention.

Detailed Summary

Two specific API settings were enabled on GPT-5.6, significantly improving its performance on the ARC-AGI-3 benchmark by tripling the scores while enhancing both reasoning capabilities and operational efficiency. The involved parties include researchers or developers who made these adjustments to the model. This improvement has broader implications for advancing artificial general intelligence (AGI) technologies and their practical applications.

Key Points

• • Enabled setting A improved reasoning capabilities.
• • Setting B enhanced data compaction significantly.
• • Combined use of both settings tripled benchmark scores.