GPT-5.6 Terra is the middle rung in OpenAI's refreshed three-tier family, sitting between the flagship Sol tier and the cost-efficient Luna tier. Independent reporting from Vorp Labs confirms the official model identifiers gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna, with pro and ultra treated as request parameters rather than separate checkpoints, so Terra inherits the same underlying lineage as its siblings while being positioned for everyday coding, reasoning, and agentic work rather than maximum capability or minimum cost.

The developer-facing shift around this release is as important as the model itself. The API replaces the old single effort scale with separate reasoning.mode (standard or pro) and reasoning.effort parameters, and caching mechanics have changed shape, with explicit breakpoints, a 30-minute minimum cache life, and 90%-off cache reads. On OpenRouter, the same upstream is reachable through OpenAI, Azure, Azure EU, and Amazon Bedrock, each with distinct latency and uptime profiles, so teams integrating Terra through OpenRouter can pick a provider based on operational priorities rather than model capability, which is a meaningful flexibility gain for production agentic pipelines.