OpenAI said it has made GPT-5.6 generally available after a limited preview, launching Sol as its top model, Terra for everyday work, Luna as its lowest-cost model, and ultra as a setting for more complex tasks.

OpenAI said Sol is its reasoning tier, Terra is a production tier, and Luna is a lightweight tier for high-volume use.

In company-reported tests, Sol scored 53.6 on Agents’ Last Exam, which measures long-running professional workflows across 55 fields.

Sol also scored 80 on the Artificial Analysis Coding Agent Index, which is built from DeepSWE, Terminal-Bench v2, and SWE-Atlas-QnA.

OpenAI said Terra performed just above Anthropic’s Claude Fable 5 on the coding index, while Luna outperformed Anthropic’s Claude Opus 4.8 at a lower estimated cost.

OpenAI also said GPT-5.6 can use Programmatic Tool Calling in the Responses API.

Its documentation said tool calling can run code and external systems in multi-step workflows, while its preview system card said the models are rated highly capable in cyber and biological or chemical risk, and include activation-based safeguards.

🔗 Source: OpenAI

Recent OpenAI developments