GPT-5.6Terra:benchmarks,costandwhereitfits

Sol and Terra share the same API family, tool surface, context size, and effort vocabulary. That gives teams a clean cascade without rewriting prompts or orchestration: Terra handles the normal path, while Sol receives tasks that fail a verifier or cross a difficulty threshold.

The price difference is large enough to matter at volume, and the current throughput observations favor Terra. The model is not merely a budget compromise; it is the operational center of the family.

Terra supports none through max reasoning. The strongest setting improves independent scores, but low and non-reasoning modes are dramatically more responsive. A well-designed router can preserve the cost advantage; a blanket max-effort setting can quietly erase it in latency and reasoning tokens.