GPT-5.6 Terra is the middle tier of OpenAI's GPT-5.6 series, which shipped on July 9, 2026 as one architecture distilled into three capability tiers: Sol, Terra, and Luna. OpenAI positions Terra as a balanced model for everyday work, and independent evaluation describes it as handling development loops, coding agents in CI/CD, and document processing up to 512K tokens. The tiering system is built around the same API surface and shared reasoning modes, with Terra trading off roughly half the token cost of the Sol flagship for modest benchmark differences.
On independent benchmarks, Terra consistently lands within three points of Sol across coding, reasoning, and agentic tests, including 77.4 on the Artificial Analysis Coding Agent Index versus Sol's 80, 63.4% on SWE-Bench Pro, and 87.4% on Terminal-Bench 2.1. Long-context recall is a clear strength over the budget Luna tier, with Terra scoring 89.6% on MRCR v2 8-needle at 256K-512K compared to Luna's 41.3%. The llm-stats composite ranks Terra sixteenth overall with a 50.8 score at a blended cost around $2.48 per the cataloged API limit, and it tops the Search and Function-Calling production benchmark. The practical fit is a mid-cost workhorse that covers most agentic coding and document workflows, reserving the more expensive Sol for production gates and security tasks where the accuracy gap widens.
