OpenAI Rolls Out GPT-5.6 to the Public — Sol, Terra and Luna

OpenAI's newest frontier model is going public today, on a schedule set partly in Washington. OpenAI is rolling out GPT-5.6 to everyone starting Thursday, July 9, moving beyond the roughly 20 vetted partners that were its only outside users during a preview that began June 26. The company says it is expanding access "globally now," across ChatGPT, the API and Codex — a staged rollout rather than a single switch thrown for everyone at once.

Three tiers, one generation

GPT-5.6 ships as three named tiers that share a generation number but advance on their own cadence. Sol is the flagship, tuned for coding, biology and cybersecurity; Terra is a lower-cost mid-tier pitched as competitive with GPT-5.5 at roughly half the price; and Luna is the fastest, cheapest option for high-volume work. Prices run $5/$30 per million input/output tokens for Sol, $2.50/$15 for Terra and $1/$6 for Luna. A reworked prompt-caching system — explicit cache breakpoints, a 30-minute minimum life and cache reads at a 90% discount — is active on all three from day one.

The benchmark wrinkle

On Terminal-Bench 2.1, Sol scores 88.8%, rising to 91.9% in a new parallel-subagent "Ultra" mode; a separate "max" reasoning setting pushes single-model effort further. The quirk is lower down the ladder: budget Luna (84.3%) actually edges mid-tier Terra (82.5%) on the test, a sign the tiers are optimized for different targets rather than a clean price-to-quality staircase. OpenAI also plans a Cerebras deployment of Sol at up to 750 tokens per second for latency-sensitive agents.

Who cleared it

GPT-5.6 spent nearly two weeks inside a review framework the Trump administration set up on June 2, with testing led by the Commerce Department's Center for AI Standards and Innovation (CAISI). Reports that the government "approved" the release drew a sharp denial: a White House spokesperson said the administration "did not give OpenAI a 'green light,' approval, or clearance," and that "decisions on timing and scope of releases rest entirely with the companies." OpenAI, for its part, said it does not believe "this kind of government access process should become the long-term default."

The field fills in

With Anthropic's Fable 5 back online since July 1 and GPT-5.6 now going public, no frontier model remains under active US government restriction for the first time since the June 12 export-control action. Google's Gemini 3.5 Pro is the lone flagship still in preview — by the company's own choice.