OpenAI unveils cheaper, more accurate GPT-6 Sol and Luna models
OpenAI announced the release of GPT-6 Sol and Luna, smaller models that cost half as much as the previous 5.6 series and claim lower error rates.
OpenAI expanded its GPT-6 lineup by launching Sol and Luna, two smaller models designed for distinct workloads. Sol is positioned for demanding programming tasks, whereas Luna focuses on repetitive, goal-driven activities like document summarization and quick question answering. The firm highlighted a 50% price cut compared with the 5.6 series, attributing the savings to advances in caching and inference efficiency.
Internal tests show Sol’s factual error rate is about half that of the earlier version, delivering Astra-level accuracy at a reduced cost. OpenAI claims both models outperform Anthropic’s top offerings, such as Fable and Opus, which released a new version shortly before the announcement. The models are being rolled out gradually across ChatGPT Work, Codex, the API, and the desktop app, with Luna also accessible to free and Go users.
Why it matters
Cheaper, more reliable AI models could broaden access and lower costs for businesses and developers.
In this story
