OpenAI has officially announced GPT-5.6 — but this isn’t the clean general-availability launch most coverage expected. The new model family, codenamed Sol, Terra, and Luna, is shipping as a government-gated limited preview, restricted to roughly 20 trusted partner organizations whose participation has been shared with the US government. The announcement lands within the predicted June 22–28 window that Polymarket traders priced at 83% odds, and follows OpenAI chief scientist Jakub Pachocki’s internal description of the model as a “meaningful improvement” over GPT-5.5.
What’s confirmed at launch
GPT-5.6 isn’t a single model — it’s a three-tier family, each named for a durable capability tier rather than a generation number:
- GPT-5.6 Sol — the new flagship, $5 input / $30 output per 1M tokens
- GPT-5.6 Terra — a lower-cost mid-tier option, $2.50 input / $15 output per 1M tokens
- GPT-5.6 Luna — the fastest, most cost-efficient tier, $1 input / $6 output per 1M tokens
Sol introduces new “max” and “ultra” reasoning effort settings — max gives the model more time to reason before responding, while ultra deploys subagents to accelerate complex, long-running work. OpenAI is also launching Sol on Cerebras hardware in July, targeting speeds up to 750 tokens per second, though that access will start limited as capacity expands.
On capability, OpenAI rates all three models as High under its Preparedness Framework for both cybersecurity and biological/chemical risk — though none cross the High threshold for AI self-improvement. In internal testing against Chromium and Firefox, Sol could identify vulnerabilities and exploitation primitives but did not autonomously chain them into a full working exploit. That capability jump is the stated reason for the staged, government-coordinated rollout rather than a standard public launch.
How GPT-5.6 compares
GPT-5.6 enters a market where Anthropic’s Claude Opus 4.8 and Google’s Gemini 3.5 Pro are the other two frontier contenders — and Gemini 3.5 Pro remains unreleased, with Google reportedly pushing it to July, even as GPT-5.6 ships its limited preview in the same window. See our full GPT-5.6 vs Claude Opus 4.8 vs Gemini 3.5 Pro comparison for the detailed breakdown. The open-weight challenger to watch in the same window is Zhipu’s GLM-5.2, which lands close to the closed-source leaders on coding at a fraction of the cost — and unlike GPT-5.6, is available to everyone right now.
Pricing and access
For now, GPT-5.6 Sol, Terra, and Luna are available only to a small group of trusted partners whose access has been approved by the US government, as part of an in-between arrangement while the Administration finalizes a formal evaluation process for frontier model releases under its cyber executive order framework. OpenAI says it expects to expand access to more companies in the coming week, with general availability targeted for “the coming weeks” after that — though no fixed GA date has been set. OpenAI has been explicit that it doesn’t see this kind of government-gated access as a long-term default, calling it a short-term step toward broader release rather than the new normal. Notably, Anthropic is reportedly going through a similar negotiation process with the government over its own next model, so this isn’t an OpenAI-specific constraint — it appears to be the new shape of frontier launches generally for the time being.
Should you switch?
For nearly everyone reading this, the honest answer right now is: you can’t switch yet — access is capped at roughly 20 approved organizations. If you’re not one of them, the practical move is to keep building on GPT-5.5, Claude, or GLM-5.2 and watch for the broader rollout in the coming weeks. If you do have preview access: Sol’s ultra mode and expanded coding/cybersecurity gains make it worth testing on long-running agentic and security-adjacent workloads, while Terra and Luna are worth benchmarking against GPT-5.5 for cost-sensitive production use before committing. We’ll update this post the moment access opens further or a firm GA date is confirmed.
Related reading
- GPT-5.6 full guide — original pre-launch coverage
- GPT-5.6 launch window analysis — the signals that called this correctly
- Gemini 3.5 Pro release date — Google’s competing model, still pending