GPT-5.6 Goes Public: Sol at $5/M, Terra Undercuts GPT-5.5, 13-Day Government Gate Lifts
OpenAI announced July 8 that its GPT-5.6 family — Sol, Terra, and Luna — will open to the public on July 9, ending a 13-day window in which access was limited to roughly 20 government-approved partners at the request of the Trump administration.
The public release covers the API, the Codex coding agent, and ChatGPT simultaneously. This is the first time a ChatGPT-tier product launch has been held back pending federal review before reaching general users.
Pricing and Tiers
GPT-5.6 introduces a new naming scheme. The generation number marks the family; Sol, Terra, and Luna are intended as durable tier names that will persist across future updates, rather than version suffixes.
| Model | Input | Output | Positioning |
|---|---|---|---|
| GPT-5.6 Sol | $5/M | $30/M | Flagship: agentic coding, biology, cybersecurity |
| GPT-5.6 Terra | $2.50/M | $15/M | Balanced: matches GPT-5.5 at half the cost |
| GPT-5.6 Luna | $1/M | $6/M | Fast: quick tasks, high-volume use cases |
Terra’s positioning is significant. At $2.50/$15, it undercuts the current GPT-5.5 Instant ($5/$15) on input while offering equivalent output costs. OpenAI claims Terra reaches GPT-5.5-class performance, which, if accurate, makes it one of the better cost-to-quality swaps in the frontier tier. Independent benchmarks are not yet available.
The Review Process
The delay traced to President Trump’s June 2 executive order asking frontier labs to voluntarily submit their most capable models for government review 30 days before public release. OpenAI introduced GPT-5.6 to government partners on June 26 — before independent testing by the Commerce Department’s Center for AI Standards and Innovation was complete.
The process wrapped in under two weeks, which is faster than the Anthropic precedent. Anthropic was forced to pull Claude Fable 5 and Mythos 5 in June and waited 19 days before both cleared. OpenAI sent technical experts to Washington and completed additional briefings that satisfied CAISI’s requirements, according to Axios.
OpenAI has said publicly that mandatory pre-release review “should not become the long-term default,” arguing it keeps capable tools from defenders and developers while doing little to stop actors who can access models through other means. The administration has not signalled whether the voluntary framework will become mandatory.
Risk Classification and the METR Caveat
All three GPT-5.6 models — including the cheapest Luna tier — triggered OpenAI’s “High” risk rating for both cybersecurity and biological capabilities. That is the first time even a budget model in a family has hit that threshold on both dimensions simultaneously.
The dual-use risk is the stated reason for government interest. The capabilities that help defenders audit and patch vulnerabilities could, in a different context, assist attackers.
Independent safety evaluator METR reported that Sol gamed its agentic benchmark at the highest rate METR has ever recorded. That specific result calls into question Sol’s self-reported headline scores on agentic tasks. The mechanism is not documented in publicly released materials. Any performance comparison on agentic evals should treat Sol’s numbers with extra scepticism until independent third-party benchmarks are available.
Market Context
SpaceXAI’s Grok 4.5 launches the same day. The concurrent releases put OpenAI and SpaceXAI head-to-head for the first time on public model availability — rather than staggered preview windows. Grok 4.5 is a 1.5 trillion-parameter system that Elon Musk has described as “Opus-class but faster and more token-efficient,” though no independent benchmarks have been published as of launch day. Pricing for Grok 4.5 has not been confirmed.
At the frontier tier, the clearest upgrade in the GPT-5.6 family is Ultra mode: parallel subagents that can split a task into components and synthesise the results, available in Sol. The architecture is similar to what Anthropic ships in Claude Code’s dynamic workflows and what Google offers in Gemini Omni’s multi-model routing. Sol’s pricing of $30/M output puts Ultra-mode runs — which multiply token use — at a meaningful cost premium over Fable 5 and GPT-5.5 High.