OpenAI Unveils GPT-6 Sol & Luna: A New Era of AI with Lower Costs & Fewer Errors

Earlier this month, OpenAI introduced GPT-6 Astra, positioning it as its most powerful and capable model to date, and touting it as the "world's best model" for a range of activities, including computer work and coding. Following this significant launch, OpenAI is now expanding the GPT-6 generation by rolling out updated versions of its smaller Sol and Luna models.

OpenAI states that while GPT-6 Astra initiated a new era of intelligence, these subsequent models aim to extend those benefits by enhancing efficiency and accessibility. The initial versions of the Sol and Luna series were first launched earlier this year, establishing different tiers within OpenAI's model hierarchy. Sol is specifically engineered for complex tasks such as coding, whereas Luna is described as more suitable for clerical work – encompassing "high-volume tasks with a clear goal, like summarizing documents, extracting information, or answering quick questions."

A key focus for OpenAI with these new releases is a substantial improvement in efficiency and affordability. Most notably, there's a significant price reduction for API access, with the 6 series models becoming available at half the cost of the 5.6 series of Sol and Luna. OpenAI attributes this price drop to advancements in caching and inference technologies.

Furthermore, OpenAI asserts that its latest models exhibit enhanced factual accuracy and boast a lower error rate when performing coding tasks. According to an internal factuality evaluation, which utilizes de-identified real-world conversations where users identified model mistakes, GPT-6 Sol reportedly makes approximately half as many errors as its predecessor. This brings its reliability to an "Astra-level" standard, but at a considerably lower cost.

In a characteristic move for OpenAI's new model introductions, the company claims that its newest offerings outperform those from its primary competitor, Anthropic. OpenAI repeatedly highlights that GPT-6 Sol and Luna manage tasks significantly better than Anthropic's top models, such as Fable and Opus. This competitive landscape was underscored by Anthropic's release of a new version of Opus 5.5 just 90 minutes prior to OpenAI's announcement, illustrating the intense rivalry between the two AI labs.

The updated versions of Sol and Luna are now accessible in ChatGPT Work and Codex for the majority of paid accounts, and via the ChatGPT API. Additionally, Luna will be made available in the desktop application and for Free and Go users. OpenAI anticipates a gradual rollout of these models to the ChatGPT app and website throughout the day.