OpenAI’s new GPT-6 Sol beats Opus 5 at less than 10% of its cost

It’s been a few weeks since OpenAI introduced GPT-6 Astra and took the internet by storm. Now, the company is introducing new, more cost-effective GPT-6 models for users who need a powerful AI model at a considerably lower price.

OpenAI announced the launch of GPT-6 Sol and GPT-6 Luna today, and the company is focused on showing how affordable and capable both models are. API pricing for both GPT-6 Sol and Luna has been reduced significantly compared to GPT-5.6 Sol and Luna. Both are now 50% cheaper to use.

The new GPT-6 models also offer better capabilities in nearly all scenarios. According to OpenAI’s testing, GPT-6 Sol outperforms models like Claude Opus 5 while being considerably cheaper. Even in more complex tasks, such as coding, GPT-6 Sol reportedly matches Claude Fable 5.1 at a much lower cost. It’s worth noting, however, that Anthropic also recently released Claude Opus 5.5, which is also faster and more affordable than Opus 5. How GPT-6 Sol compares with Opus 5.5 remains to be seen.

More importantly, OpenAI has further improved its models’ accuracy. The company claims both GPT-6 Sol and Luna make noticeably fewer mistakes than their predecessors. According to internal test data, GPT-6 Sol makes half as many mistakes as its predecessor, GPT-5.6 Sol. Meanwhile, GPT-6 Luna is also much better at getting facts right, though it seems to perform best at higher effort levels.

GPT-6 Sol and Luna are available starting today in ChatGPT Work and Codex to all ChatGPT Pro, Plus, Business, Enterprise, and Edu users. Users on the ChatGPT Free and Go plans can also access GPT-6 Luna in the desktop app. That said, OpenAI is gradually rolling out the new models, so you may have to wait a bit before you can access them.