GPT-6 Sol and Luna halve prices with limited performance gains
OpenAI has introduced GPT-6 Sol and GPT-6 Luna as lower-cost successors to GPT-5.6 Sol and Luna. Their token prices are cut by 50 percent: Sol costs $2 per million input tokens and $10 per million output tokens, while Luna costs $0.10 and $0.50. OpenAI attributes the reductions to improvements in caching and inference. Terra, previously the cheapest model, is no longer available.
OpenAI describes Sol as a model for complex recurring work such as coding, debugging and data analysis. Luna targets high-volume, structured tasks including summarization, information extraction and short answers. Both models offer a 90 percent discount on cached input tokens, alongside new caching diagnostics. They are available in the API as gpt-6-sol and gpt-6-luna, with gradual ChatGPT access for eligible subscribers.
OpenAI’s comparisons emphasize cost efficiency against Anthropic models. Artificial Analysis found that per-task costs fell by about half, but intelligence scores remained close to GPT-5.6 levels: Sol rose from 47 to 48 points, while Luna stayed at 37. The analysis also reported declines on GDPval-AA v2.1, with lower presentation quality and incomplete results contributing to the regressions. The models’ main distinction is therefore lower cost rather than a major increase in measured capability.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.