gpt-5.6-terra

Overview. GPT-5.6 Terra is OpenAI's mid-tier ('balanced') model in the GPT-5.6 lineup, generally available since July 2026. It pairs a 1.05-million-token context window with a 128K max output and accepts text, image, and file input while returning text. Developers reach for Terra when they want a general-purpose Inference option that balances intelligence and cost without stepping up to a dedicated reasoning-first variant.

Key strengths. Independent Artificial Analysis benchmarking puts Terra's coding index at 76.7, its strongest measured dimension, alongside a general intelligence index of 56.6. It supports a configurable reasoning-effort control (none through max, default medium), letting a request dial up deliberation only when a task needs it. Context caching is available, which helps repeated-prompt workloads reuse earlier context efficiently.

When to use it. Terra fits medium-complexity, interactive workloads: coding assistance, document analysis over long inputs, and general chat or assistant use where a fast, balanced model is preferable to a heavier reasoning-first tier. Its 1.05M-token context window and native file input make it a reasonable choice for tasks that require ingesting large documents or codebases in a single call. Because its measured agentic index (50.2) trails its coding score, it's a better fit for single-turn or lightly agentic tasks than for long multi-step autonomous workflows.

Limitations. Terra is text-and-image-in, text-out only — it has no audio input or output. It's a closed-weight, proprietary model with a training cutoff of February 16, 2026, so it won't have knowledge of events after that date. As the family's balanced mid-tier, it sits below OpenAI's dedicated reasoning-first variant on complex reasoning tasks.

Sovereignty note. GPT-5.6 Terra is built and hosted by OpenAI, a US-based creator, and access is governed by OpenAI's standard API Terms of Use. No EU data-residency or CLOUD Act exposure details were available in research for this model, so it should not be treated as a Sovereign option without separate verification.