GPT-6 received Sol and Luna: Terra remains a model of the GPT-5 line. 6

OpenAI released two GPT-6 models for the API on September 22: GPT-6 Sol and GPT-6 Luna. They take text and images, return text, and are available through the Responses API and Chat Completions API. There is an important limitation in the name of the line: Terra refers to GPT-5.6, not GPT-6.

GPT-6 Sol OpenAI is positioned for complex development and agency scenarios. The GPT-6 Luna is designed for focused, repetitive tasks with a high volume of queries. Both models have a contextual window of 1.05 million tokens, the maximum response volume is 128,000 tokens; reasoning levels from none to max are available.

For requests with input contexts of up to 272,000 tokens, the standard price for GPT-6 Sol is $2 per million input tokens and $10 per million weekends. The GPT-6 Luna has $0.10 and $0.50 respectively. The difference is important for product teams: Luna is suitable for mass-scale operations of the same type, such as initial classification of applications, draft signatures to images or checking the structure of content. Sol is worth choosing for scenarios where an agent has to work with multiple tools, code, or ambiguous tasks. This is a recommendation on the intended designation of models, not the result of an independent comparative test.

The name Terra is retained in the previous GPT-5.6 family. In it, the gpt-5.6-sol remains the flagship model, the gpt-5.6-terra balances quality and price, and the gpt-5.6-luna is designed for economical mass requests. In the current GPT-6 family, OpenAI lists Astra, Sol and Luna. Therefore, the gpt-6-terra identifier cannot be incorporated into the integration: the official directory does not specify it.

For design and content processes, this separation helps avoid two mistakes. Do not send a complex user flow audit to a cheap model without checking the quality of real layouts and data. And you don’t have to default to Sol for every automatic action: in a regular stream of simple tasks, the cost of Luna will be noticeably lower. Before starting, it is useful to run the same set of working cases through both models, evaluate the accuracy, response time and price of the full scenario, then fix the model to a specific stage of the process.

OpenAI also warns that the availability of models, tools, and reasoning settings depends on the product and version. This news describes API models; the presence of the same-name modes in a specific tariff ChatGPT should be checked separately in its catalog of models.

Sources: OpenAI API Changelog — релиз GPT-6 Sol и GPT-6 Luna, 22 сентября 2026; OpenAI Model guidance — семейство GPT-6 и особенности интеграции; OpenAI Model guidance — семейство GPT-5.6, включая Terra.