OpenAI launches GPT-5.6 with Sol, Terra and Luna tiers across ChatGPT, API and developer platforms

OpenAI has released GPT-5.6, a new model family offered in three tiers — Sol, Terra and Luna — across ChatGPT, Codex and the OpenAI API.

Three tiers for different workloads

In its product page for GPT-5.6, OpenAI describes the release as available across ChatGPT, Codex and the API, with the model family split into Sol, Terra and Luna. OpenAI’s published materials include API pricing, context and caching details, and benchmark tables covering areas such as coding, science, multimodal tasks, long-context work and cybersecurity.

Amazon Web Services also said OpenAI GPT-5.6 Sol, Terra and Luna are generally available on Amazon Bedrock. In AWS’s description, Sol is positioned as the flagship reasoning model, Terra as a balance of performance and efficiency, and Luna as a faster, more cost-efficient option for inference.

That tiering is echoed by Axios, which reported OpenAI’s GPT-5.6 release and described Sol as the most powerful tier, Terra as a middle option balancing efficiency and capability, and Luna as aimed at speed and affordability.

Developer platforms add support

Vercel’s AI Gateway changelog lists GPT-5.6 Sol, Terra and Luna as available model slugs. Vercel summarizes the family around agentic work, coding, biology, cybersecurity and pricing-tier choices, giving developers another route to access the models through its gateway product.

The availability across OpenAI’s own API, Amazon Bedrock and Vercel’s AI Gateway suggests OpenAI is distributing GPT-5.6 through both first-party and third-party infrastructure channels. For enterprise and developer teams, the practical distinction may be less about a single model name and more about which tier fits a workload’s latency, cost and capability requirements.

Benchmarks and pricing are central to the launch

OpenAI’s own GPT-5.6 page says the company has published benchmark tables for coding, science, multimodal, long-context and cybersecurity tasks. Those categories point to the markets OpenAI is targeting with the release: software development, technical research, security analysis and applications that require larger context windows or multiple input types.

The company also lists API pricing and caching information, according to its product page. Those details matter because model selection increasingly depends not only on raw benchmark scores but also on operating cost, response latency and whether repeated context can be cached efficiently.

AWS’s positioning of Luna as fast and cost-efficient, Terra as balanced, and Sol as the flagship reasoning option reflects that same trade-off. Rather than presenting GPT-5.6 as a single general-purpose product, OpenAI and its platform partners are framing it as a family of models for different usage patterns.

What is confirmed — and what is not

The confirmed facts from the cited sources are that OpenAI has introduced GPT-5.6; that the family includes Sol, Terra and Luna; that OpenAI says it is available across ChatGPT, Codex and the API; and that AWS and Vercel list the models as available through Amazon Bedrock and AI Gateway, respectively.

The available source excerpts do not provide enough detail to independently evaluate OpenAI’s benchmark claims or compare GPT-5.6 against rival frontier models. They also do not establish real-world performance outside the tests and product descriptions published by OpenAI and its distribution partners.

For now, the release is best understood as a tiered expansion of OpenAI’s model lineup, with OpenAI, AWS and Vercel all presenting GPT-5.6 as a family designed to let users choose between higher-end reasoning, balanced performance and lower-cost speed.