OpenAI Launches GPT-6 Sol and Luna… API Prices Cut by 50% Compared to GPT-5.6

OpenAI has unveiled new models GPT-6 Sol and GPT-6 Luna, which extend the intelligence advantages of GPT-6 Astra in a cost-efficient manner. Through improvements in caching and inference infrastructure, API costs have been reduced by 50% compared to the previous GPT-5.6 promotional pricing, while GPT-6 Astra remains the top-performing model.

Performance and Cost Efficiency

GPT-6 Sol demonstrates strong performance in specialized tasks and coding. On the AutomationBench evaluation, it scores higher than Claude Opus 5 while costing only about 9%. On DeepSWE v1.1, it achieves a score of 68.8%, showing performance similar to Claude Fable 5 xhigh while reducing costs by approximately 80%. GPT-6 Luna achieves performance similar to GPT-5.6 Sol at higher effort levels for about 1/100th of the cost, recording 66.6% on DeepSWE v1.1 and offering costs 93–96% lower than competing models.

In terms of factuality, GPT-6 Sol has seen errors halved compared to its predecessor, approaching the reliability level of Astra. In computer use capabilities (OSWorld 2.0), GPT-6 Sol shows performance similar to Claude Opus 5 while reducing costs by 80%.

Caching Improvements and Availability

Caching capabilities have been significantly enhanced for agent and long-conversation usage. The cost of reading cached input tokens is discounted by 90%, and the Prompt Caching Dashboard and diagnostics tool allow users to monitor cache status and analyze causes of misses. In GitHub cases, the proportion of prompt token processing decreased by more than 50%, improving Copilot response speeds.

Both models are immediately available to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. Free and Go users can access only Luna via the desktop app, and the API model names are gpt-6-sol and gpt-6-luna.