OpenAI has released updated versions of its Sol and Luna models, marking the next step in the GPT-6 generation following the earlier launch of GPT-6 Astra.
The company states that while Astra introduced a new level of intelligence, these smaller models extend those benefits by making the technology more efficient and accessible.
The model hierarchy
Sol and Luna were first introduced earlier this year as different tiers within OpenAI’s model range. Sol handles complex work like coding. Luna is intended for clerical tasks, such as summarising documents, extracting information, or answering quick questions.
OpenAI is emphasising significant improvements in efficiency and affordability. API access for the GPT-6 series will cost half as much as the GPT-5.6 series. The company attributes this price drop to improvements in caching and inference.
The new models also claim to produce answers with higher factual accuracy. They show a lower error rate when writing code.
According to the announcement, internal factuality evaluations show GPT-6 Sol makes about half as many mistakes as its predecessor. It reaches Astra-level reliability at a much lower cost.
Competition heats up
As is typical with OpenAI releases, the company claims its new offerings outperform those of its primary competitor, Anthropic. OpenAI repeatedly states that GPT-6 Sol and Luna handle tasks substantially better than Anthropic’s top models, including Fable and Opus.
Anthropic released a new version of Opus 5.5 just 90 minutes before OpenAI’s announcement. This timing reflects the intense competition between the two firms.
Availability
The new versions of Sol and Luna are available now in ChatGPT Work and Codex for most paid accounts. They are also available in the ChatGPT API. Luna will be available in the desktop app and for Free and Go users.
OpenAI expects to roll out the models to ChatGPT, the app and website, gradually throughout the day.
What it means
Developers and businesses using API calls for coding or document processing can expect to pay less while seeing fewer hallucinations and errors in the output.
