OpenAI Announces GPT-5.6 Sol, Terra, and Luna in Limited Preview

OpenAI has officially announced the GPT-5.6 family, introducing three new AI models designed for different performance and pricing tiers. The lineup includes GPT-5.6 Sol, Terra, and Luna, with each model targeting different workloads and budgets.

Unlike previous launches, GPT-5.6 is entering a limited preview available only to a small group of trusted partners. OpenAI said the restricted rollout follows discussions with the U.S. government, which requested that the initial release be limited to selected organizations. The company added that it does not believe this approval process should become the long-term standard for future AI releases.

GPT-5.6 introduces three new models

GPT-5.6 Sol is OpenAI’s flagship model, built for demanding reasoning, coding, scientific research, cybersecurity, and agentic workloads.

GPT-5.6 Terra targets everyday professional use, offering performance comparable to GPT-5.5 while costing half as much.

GPT-5.6 Luna is the fastest and most affordable option in the lineup, providing solid performance at the company’s lowest price point.

OpenAI plans to make all three models broadly available across ChatGPT, Codex, and its API over the coming weeks.

Stronger reasoning and agent performance

According to OpenAI, GPT-5.6 Sol is its most capable model to date.

The company says the model delivers improved agentic performance across coding, biology, and cybersecurity tasks. On the Terminal-Bench 2.1 benchmark, GPT-5.6 Sol achieved a score of 91.9%, setting a new record and outperforming Anthropic’s Claude Mythos 5.

GPT-5.6 also introduces a new “max” reasoning effort for more complex problems. For even larger workloads, OpenAI added an “ultra” mode that uses multiple subagents to complete tasks more efficiently than a single AI agent.

Improved safety and predictable caching

OpenAI said GPT-5.6 Sol includes a strengthened safety system with better protections against higher-risk requests, sensitive cybersecurity prompts, and repeated misuse.

The company also spent several weeks pressure-testing the model against real-world attacks before launch.

Developers will also benefit from improved prompt caching. GPT-5.6 supports explicit cache breakpoints and guarantees a minimum cache lifetime of 30 minutes, making API usage more predictable.

Pricing and availability

OpenAI announced the following API pricing:

• GPT-5.6 Sol: $5 per million input tokens and $30 per million output tokens.
• GPT-5.6 Terra: $2.50 per million input tokens and $15 per million output tokens.
• GPT-5.6 Luna: $1 per million input tokens and $6 per million output tokens.

OpenAI also confirmed that GPT-5.6 Sol will launch on Cerebras in July, offering processing speeds of up to 750 tokens per second for select customers.

While GPT-5.6 is currently limited to trusted partners, OpenAI says Sol, Terra, and Luna will become broadly available in ChatGPT, Codex, and the API in the coming weeks.