OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes

OpenAI has expanded its GPT-6 lineup with refreshed Sol and Luna models, positioning them as more efficient, lower-cost variants that extend capabilities introduced by GPT-6 Astra. Astra debuted earlier in the month as the lab’s highest-capability model; the new Sol and Luna updates aim to bring aspects of that performance into smaller, task-specific models. Sol and Luna occupy different spots in OpenAI’s product hierarchy: Sol is intended for more complex work such as programming, while Luna is targeted at high-volume clerical tasks like document summarization, information extraction, and quick question answering. OpenAI says the updated models achieve substantial gains in efficiency and factual accuracy compared with the previous 5.6-series versions. A headline change is the pricing: OpenAI will offer the 6-series Sol and Luna on the API at roughly half the cost of the 5.6-series equivalents, a reduction the company credits to improved caching and inference. Internally measured factuality metrics also show improvements; OpenAI reports GPT-6 Sol makes about half as many mistakes on a de-identified real-world conversational evaluation where users had previously flagged errors. OpenAI contrasted the new releases with competing products from Anthropic, asserting that Sol and Luna outperform Anthropic’s top models such as Fable and Opus. The launch came shortly after Anthropic pushed an update to Opus 5.5 about 90 minutes earlier, underscoring quick back-and-forth product activity between the firms. The updated Sol and Luna models are being rolled out across OpenAI’s paid offerings and developer tools: they are available in ChatGPT Work and Codex for most paid accounts and accessible via the ChatGPT API. Luna will also appear in the desktop ChatGPT app and be available to Free and Go users, with a gradual rollout to the broader ChatGPT app and website expected during the day.