OpenAI expanded the GPT-6 family on September 22, 2026, with updated Sol and Luna models offering lower costs and fewer errors. The company said these models make the capabilities of GPT-6 Astra, which it had previously introduced and described as “the world’s best model,” more accessible.
Of the models introduced at the beginning of the year, Sol was designed for complex tasks such as coding, while Luna was developed for high-volume office work such as document summarization, information extraction, and answering quick questions. The GPT-6 series’ Sol and Luna models will be offered at half the price of the 5.6 series, thanks to caching and inference improvements.
OpenAI announced that GPT-6 Sol makes approximately half as many errors as the previous model, with a particularly lower error rate in coding, and achieves Astra-level reliability at a lower cost. The company also argued that Sol and Luna outperform models such as Anthropic’s Fable and Opus. Anthropic’s release of a new version of Opus 5.5 just 90 minutes before OpenAI’s announcement highlighted the competition between the two companies.
The new versions of Sol and Luna have been made available in ChatGPT Work and Codex for most paid accounts, as well as through the ChatGPT API. Luna will also be offered to Free and Go users through the desktop application. OpenAI plans to roll out the models gradually throughout the day to ChatGPT’s app and website.
Why it matters
The main consequence of this update is that advanced AI capabilities are being brought to a broader range of users and workflows at a lower cost of use. Teams that write code and employees who carry out intensive document and information processing will be able to benefit from models differentiated according to different needs; the scope of access between free and paid accounts will therefore also become important. When the pricing advantage is considered alongside OpenAI’s performance claims, it could shape model selection based not only on technical capacity but also on cost. However, Sol and Luna’s superiority over their competitors and previous models is conveyed in the article through company statements rather than independent tests; for this reason, real-world results and the scope of the gradual rollout remain among the open questions.
Background
OpenAI is not a new name in the FikirPilot archive: over the last 90 days, we have published 83 articles mentioning the name; the most recent is dated October 3, 2026.