In early September, OpenAI officially released Astra, the first GPT-6 model ever. Just a few weeks later, Sam Altman’s company has begun rolling out two new, more specialized models: GPT-6 Sol and GPT-6 Luna.
These are two new usage modes for the world’s most popular chatbot, designed to make solving tasks and answering queries faster and more straightforward. Users simply select their preferred model, enter a prompt in the chat bar, and wait for the AI agent to deliver a clear, thorough answer, organizing its workflow to meet deadlines along the way.
What are GPT-6 Sol and GPT-6 Luna?
According to the release notes for the new version of the ChatGPT desktop app, GPT-6 Sol and GPT-6 Luna use fewer tokens to run than the earlier GPT-5.6 and GPT-6 Astra models. Sol is built for complex programming and agent-based workflows, while Luna is recommended for high-volume, narrowly targeted tasks.
Both new models are currently rolling out to Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access Luna too, but only through the desktop app. For now, OpenAI has not released a “Terra” model for the GPT-6 family, despite earlier rumors that one was coming.
OpenAI appears to be shelving its planned mid-sized model in favor of the three it has already shipped. If GPT-6 Astra introduced a new generation of artificial intelligence, these two new models are meant to help spread its benefits more broadly, pushing the limits of cost efficiency — in OpenAI’s own words from its official launch announcement.
The advantages of the new models
GPT-6 Sol and GPT-6 Luna are arguably OpenAI’s best models yet in terms of value for money, since they draw on all of Astra’s power and improvements while using fewer tokens per operation. The goal is for both models to combine strong coding performance with lower API prices, giving developers more room to build.
For everyday computer use, Astra remains OpenAI’s strongest model overall. Sol and Luna, however, offer more cost-effective performance, including against rival models. On OSWorld 2.0, a benchmark that tests AI models on real computer-use tasks, Sol running at a high reasoning-effort setting matches the score of Claude Opus 5 running at a medium effort setting, while costing roughly 80% less per task.
Editor’s note
This article was originally published in Italian on money.it by Pasquale Conte on September 24, 2026 as «OpenAI migliora ChatGPT con altri due modelli GPT-6. Ecco le novità». It has been translated and adapted for an international audience by the Money.it International desk.
