OpenAI Releases GPT-6 Sol and Luna at Half the Price of Previous Versions

OpenAI has unveiled new models in the GPT-6 family—Sol and Luna—whose main feature is a substantially lower token cost than the flagship Astra. The developers trained the new models using the same methods as Astra, with an emphasis on significantly improving efficiency and affordability—primarily by reducing the cost of API access.

The new versions will cost half as much as their GPT-5.6-based predecessors, which is attributed to optimized caching and inference processes. GPT-6 Sol will cost $2.00 per million input tokens and $10.00 per million output tokens, while GPT-6 Luna will cost $0.10 per million input tokens and $0.50 per million output tokens.

The benchmarks provided by the company show that, on average, the new AI models perform slightly better on tasks than the GPT-5.6 versions, while processing requests costs considerably less. For example, in AutomationBench, GPT-6 Luna at maximum effort scored 3% higher than GPT-5.6 Luna while costing half as much.

OpenAI also noted that the new neural networks reason more effectively and make factual errors significantly less often. When mistakes do slip through, Sol and Luna are more likely to detect them on their own.

The new versions of the Sol and Luna AI models are already available in ChatGPT Work and Codex for most paid subscribers, as well as through the ChatGPT API. Luna will also soon appear in the desktop app and become available to users on the free and Go plans.