OpenAI released information on how two API settings improved GPT-5.6 performance on ARC-AGI-3. The implementation led to boosting scores and efficiency on the benchmark. These configuration adjustments offer a streamlined approach to optimizing model outputs.
The operational improvements were achieved by retaining reasoning during task processing. In addition, enabling compaction further supported these performance gains across the test suite. Together, these two mechanisms directly contributed to boosting scores and efficiency.
The findings illustrate how target configuration choices influence model output on evaluation benchmarks. The report details how two API settings improved GPT-5.6 performance on ARC-AGI-3 tasks. Overall, the update highlights the technical benefits of retaining reasoning and enabling compaction.