OpenAI's three-tier GPT-5.6 model family — Sol, Terra, and Luna — became publicly available on July 9, 2026, across ChatGPT, the API, and Codex, ending a 12-day voluntary government review period that made this the most closely scrutinized AI model launch in OpenAI's history.
The Government Review That Preceded the Launch
Before the public rollout, the Trump administration's AI cybersecurity directive asked OpenAI to hold the models for review. The Department of Commerce's Center for AI Standards and Innovation (CAISI) conducted additional safety testing and held direct technical meetings with OpenAI engineers over that 12-day window. The testing covered disallowed content, vision capabilities, destructive action risks, cybersecurity applications, and biological safeguards.
The White House subsequently issued a statement saying "no formal permission or clearance was required," positioning the process as voluntary cooperation rather than a regulatory gate. Still, OpenAI acknowledged the arrangement, stating: "We don't believe this kind of government access process should become the long-term default," while confirming compliance to accelerate the path to public availability.
Three Models, Three Use Cases
The GPT-5.6 family is structured as a clear three-tier offering:
Sol is the flagship model, built for complex reasoning, advanced coding, and extended agentic workflows. It introduces a "max reasoning effort" mode and an "ultra mode" that deploys subagents for intricate multi-step tasks. OpenAI validated Sol on Terminal-Bench 2.1 for command-line operations, ExploitBench for cybersecurity scenarios, and SecureBio for biology-focused evaluations. Pricing is $5 per million input tokens and $30 per million output tokens.
Terra targets everyday professional use, delivering performance comparable to GPT-5.5 at roughly half the cost. At $2.50 per million input tokens and $15 per million output tokens, it is positioned as the workhorse for teams running high-volume production workloads where frontier capability is needed without flagship pricing.
Luna is the lightweight, low-latency tier designed for speed-sensitive applications. At $1 per million input tokens and $6 per million output tokens, Luna covers use cases where response time matters more than maximum reasoning depth.
Key Highlights
- Full public availability started July 9 across ChatGPT, API, and Codex
- Sol introduces "ultra mode" with subagent orchestration for long-horizon tasks
- Terra delivers GPT-5.5-class performance at 50% lower cost
- Luna targets latency-sensitive, high-throughput applications
- Safety testing addressed cybersecurity, biology, vision, and destructive-action risks
- 12-day voluntary review by CAISI preceded the public launch
- OpenAI resists making government review a permanent release requirement
What This Means for Developers
The GPT-5.6 family is the first OpenAI release explicitly tested against government cybersecurity benchmarks before going live. Sol's ExploitBench and SecureBio evaluations signal that OpenAI is investing in domain-specific safety validation beyond generic harm filters, a meaningful shift for enterprise and government customers evaluating AI risk.
The three-tier pricing structure also simplifies deployment decisions: teams can route light tasks to Luna, core workloads to Terra, and reserve Sol for the highest-complexity reasoning and agentic pipelines — all within a single model family with a consistent API surface.
Background
GPT-5.6 was first unveiled to a limited set of government-vetted preview partners in late June 2026. The voluntary 30-day advance-notice requirement — part of the Trump administration's AI cybersecurity order — was interpreted by OpenAI as requiring a shorter embargo period before broader release. The White House voluntary AI framework, announced in early July, provides broader guidance for how frontier model developers should engage with government review processes going forward.
What's Next
The staged rollout will continue expanding across ChatGPT subscription tiers and enterprise accounts over the coming weeks. OpenAI has signaled that further models in the GPT-5.6 line — including fine-tuning access and dedicated deployment options — will follow later this summer.
Source: Engadget