OpenAI cuts API prices in half for GPT-6 Sol, Luna models

You'd be forgiven for thinking OpenAI was watching the clock. Just 90 minutes after Anthropic rolled out an upgraded Opus 5.5, OpenAI dropped GPT-6 Sol and Luna, the latest entries in a model hierarchy that started with GPT-5.6 back in July.

The rivalry between the two labs has rarely looked this synchronised or this deliberate.

OpenAI GPT-6 Sol and Luna models

Sol is built for heavy lifting: coding and complex, multi-step tasks. Luna is the opposite end of the spectrum, tuned for high-volume, low-complexity work summarising documents, pulling out data points, and answering quick questions.

OpenAI frames the pairing as intelligence "extended" from GPT-6 Astra, its top-tier release, rather than a replacement for it.

While API access to the 6 series costs half as much as what it cost with the 5.6 series, OpenAI claims this reduction is due to improved caching and inference efficiency. For software developers who are building applications using these models, the price matters more than any benchmark score.

It is said that GPT-6 Sol reduces factual errors by 50% compared to its predecessor on an internally developed benchmark based on user-reported errors. The company is also highlighting a reduced coding error rate, but as it often happens with self-reported benchmarking, it will take some time before independent verification is obtained.

But that is not all; OpenAI went further to claim that GPT-6 Sol outperforms Anthropic's Fable and Opus models in several aspects. It was just a few hours ago that Anthropic released its own version, Opus 5.5, at lower prices and higher speed.

Sol and Luna are live today in ChatGPT Work, Codex, and the API for most paid accounts, with Luna also reaching the desktop app and Free and Go tier users.