OpenAI releases GPT-6 Sol and Luna models with reduced pricing and improved accuracy
OpenAI has introduced updated versions of its GPT-6 Sol and Luna models, which are designed for complex coding and clerical tasks respectively. The new models are available at half the API cost of their predecessors, thanks to improvements in caching and inference, and they reportedly make about half as many factual errors. OpenAI claims they outperform Anthropic's top models, and they are rolling out to ChatGPT and Codex for paid users.
The Sol and Luna models occupy distinct roles within OpenAI's hierarchy, with Sol aimed at complex coding scenarios and Luna suited for high-volume clerical work such as document summarization and quick information retrieval. The price reduction stems from engineering advances in caching and inference efficiency rather than reduced capability.
OpenAI's internal factuality evaluations, based on de-identified user conversations where mistakes were flagged, indicate the new Sol model commits roughly half the errors of its predecessor. The timing is notable — Anthropic shipped an updated Opus 5.5 just 90 minutes before this release, underscoring the competitive pressure between the two labs.
The reduced pricing and improved accuracy could significantly lower barriers for businesses integrating AI into routine workflows, particularly for high-volume clerical tasks where cost has been a limiting factor. As coding and summarization tools become more reliable and affordable, smaller organizations may adopt them more readily, potentially reshaping job functions in administrative and software development roles. However, the competitive claims between OpenAI and Anthropic remain unverified independently, and users should weigh performance assertions cautiously.