GPT-6 Sol targets complex but repeatable work (code review, feature development, data analysis), while GPT-6 Luna handles high-volume, well-defined tasks such as document summarisation, information extraction and quick queries. Both models are trained using methods similar to GPT-6 Astra, the company's top-tier reasoning model.
The price cut accompanies caching and inference improvements that OpenAI says reduce its own serving costs. GPT-5.6 Luna usage grew more than 10-fold after an 80% price reduction in July, which OpenAI cites as evidence that lower prices drive adoption. Replit attributed similar cuts to its decision to open free access to millions of users.
Both new models score better than their GPT-5.6 predecessors on OpenAI's internal alignment evaluations, including lower rates of misleading claims in coding outputs. OpenAI describes those evaluations as covering challenging scenarios rather than typical use, and says a full system card will follow within 24 hours. The GPT-6 Astra model continues to serve the most demanding workloads; Databricks deployed it to around 3,500 engineers after internal tests against their previous leading models on system design tasks.
The three-model GPT-6 lineup now covers different cost and complexity points: Astra for the hardest problems, Sol for regular complex work, Luna for scale.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
