Just weeks after introducing GPT-6 Astra as its premier model for coding and computer tasks, OpenAI has broadened the GPT-6 family with updated iterations of its smaller Sol and Luna models. The company describes the new releases as an effort to make the advanced intelligence introduced by Astra more efficient and widely accessible.
Sol continues to target complex workflows such as software development, while Luna remains focused on high-volume clerical duties like document summarization, information extraction, and rapid Q&A. OpenAI highlights that these updates bring substantial gains in both affordability and performance.
A key selling point is a 50% reduction in API pricing compared to the previous 5.6 generation, a move the company attributes to advancements in caching and inference technologies. Additionally, OpenAI reports a marked decrease in factual errors, claiming that GPT-6 Sol produces roughly half the mistakes found in its predecessor based on internal evaluations of real-world user conversations.
In what is standard for its launch cycles, OpenAI asserted that its latest models outperform Anthropic’s flagship offerings, specifically naming Fable and Opus. The release follows a tight race, with Anthropic dropping a new Opus 5.5 version just 90 minutes prior to OpenAI’s announcement.
The updated models are currently accessible via ChatGPT Work and Codex for most paid subscribers, as well as through the API. Luna will also reach users on the Free and Go tiers within the desktop application. OpenAI plans a gradual rollout across the broader ChatGPT web and mobile platforms throughout the day.
Leave a Reply