OpenAI releases GPT-6 Sol and Luna, API prices cut in half

OpenAI has released GPT-6 Sol and GPT-6 Luna, further expanding the new-generation model series that began this month with GPT-6 Astra. Sol is positioned for complex programming and agent workflows, while Luna is primarily aimed at high-frequency tasks such as document summarization, information extraction, and rapid Q&A. OpenAI stated that, thanks to improvements in caching and inference efficiency, the API prices for the two models are 50% lower than the promotional pricing of the GPT-5.6 series. Under standard rates, Sol's input, cached input, and output prices per million tokens are $2, $0.20, and $10, respectively; Luna's are $0.10, $0.01, and $0.50, respectively. In terms of performance, OpenAI said the new models show improvements in factual accuracy and programming reliability. In internal evaluations built on anonymous real-world conversations, GPT-6 Sol produced roughly half the errors of the previous generation and achieved reliability close to Astra at a lower cost. The company also released test results claiming that Sol and Luna outperform Anthropic's Fable and Opus series on multiple tasks, though these conclusions are primarily based on OpenAI's internal evaluations. Both models are now available via API and are gradually rolling out to Plus, Pro, Business, Enterprise, and Edu users for ChatGPT Work and Codex; Free and Go users can use Luna in the desktop app.