Guide for GPT-5.6 Builders

The GPT-5.6 model family delivers frontier-level agent performance at radically lower costs, raising the bar for agent capabilities. Developers can build faster and more capable agents by choosing smarter models and leveraging new API control features for reasoning continuity, multi-agent orchestration, and programmatic tool calling.

Previously, using high-performance flagship models was common for handling long-horizon tasks, but in the GPT-5.6 family, the Luna and Terra models offer performance comparable to flagship models thanks to their high cost efficiency. For example, the Luna model maintains similar extraction accuracy at approximately 1/18th the cost of GPT-5.5, significantly reducing expenses for browser agent tasks.

Key improvements include:

• Cost and Performance Optimization: GPT-5.6 Sol outperforms previous generations' high reasoning settings even with low reasoning effort.
• Strategic Model Selection: Use Terra or Luna for bulk data extraction or repetitive steps to optimize costs, and reserve flagship models only for critical tasks requiring high performance.
• Responses API Enhancements: New features such as reasoning persistence, which maintains reasoning across model turns, improve agent operational efficiency.