OpenAI previews GPT-5.6 Sol, Terra, and Luna with new reasoning tiers

OpenAI began a limited preview of the GPT-5.6 family on June 26. The family introduced Sol as the flagship model, Terra as a balanced option, and Luna as the fastest and least expensive tier. Initial access was limited to selected organizations through the API and Codex, with broader availability planned after the preview.

Reasoning becomes an explicit product dimension

GPT-5.6 added a max reasoning effort for Sol and an ultra mode that uses subagents for complex work. OpenAI highlighted coding, biology, and cybersecurity evaluations and said Sol set a new result on Terminal-Bench 2.1. The preview also introduced more predictable prompt caching, including explicit cache breakpoints and a minimum cache lifetime. These capabilities change the shape of an integration: model tier, reasoning effort, agent topology, and cache policy can each affect task quality, latency, and cost.

The release paired stronger cyber capabilities with layered safeguards. OpenAI described model refusals, streaming misuse classifiers, account-level review, differentiated access, and ongoing automated and human red teaming. It assessed the models below the Cyber Critical threshold at preview time while acknowledging that benchmarks cannot cover every deployment or tool combination.

Preview status should shape deployment

Limited access is a signal to keep workloads reversible. Teams should capture model snapshots, reasoning settings, cache configuration, tool permissions, and evaluator versions for every trial. Compare complete-task success and review burden across Sol, Terra, and Luna rather than assigning tiers solely by prompt difficulty.

For security work, test both allowed defensive workflows and prohibited boundaries with authorized evaluation cases. Additional reasoning or subagents expand the action surface, so approval scopes and budgets must apply to the whole run. The preview offered an early technical contract; production commitments should wait for the later availability and documentation that match the intended environment.