The new GPT-5.6 model family is now available, featuring three tiers: Sol, the flagship model, Terra for general tasks, and Luna for cost efficiency. GPT-5.6 Sol advances AI capabilities, setting new benchmarks in coding, cybersecurity, and scientific reasoning. It demonstrates superior performance per dollar, achieving comparable or better results than previous models with fewer tokens and lower costs.

GPT-5.6 introduces ultra, a mode that orchestrates multiple agents in parallel to accelerate complex tasks. This multi-agent approach improves speed and output quality on demanding workloads. Sol also shows enhanced design judgment, creating polished interfaces and refining rendered results. The models integrate with professional workflows, processing data from sources like Slack and Google Drive to generate expert-level artifacts, including editable presentations and refined documents.

In coding, GPT-5.6 Sol leads with an 80.2 score on the Artificial Analysis Coding Agent Index, outperforming competitors while using fewer tokens and less time. The family also excels in cybersecurity, with Sol scoring 73.5% on Exploit Bench1, a significant leap from GPT-5.5. The new models incorporate enhanced safety measures, tested through extensive human and automated red teaming, to prevent misuse without hindering legitimate applications.