Just weeks after rolling out its flagship GPT-6 Astra model - which the company positioned as the most capable AI system it has ever built and claimed the title of the world's leading model for tasks ranging from software development to general computer use - OpenAI is broadening the GPT-6 lineup with refreshed iterations of two smaller models: Sol and Luna.

In an official announcement, OpenAI described the expansion by saying that while GPT-6 Astra established a new benchmark for intelligence, the Sol and Luna updates are intended to make that level of capability more efficient and widely accessible to a broader range of users and developers.

What Sol and Luna Are Designed to Do

Sol and Luna were first introduced earlier this year as part of OpenAI's tiered model architecture, with each serving a distinct purpose. Sol is built to handle demanding, technically complex tasks - most notably coding and software engineering workflows. Luna, on the other hand, is aimed at higher-volume, more routine operations. OpenAI describes Luna as well-suited for clerical and administrative use cases, such as summarizing documents, pulling out specific pieces of information from larger texts, or responding to straightforward questions quickly and at scale.

The updated versions of both models carry forward the same general positioning but come with meaningful improvements across several dimensions, including accuracy, cost, and overall reliability.

Significant Price Reductions for API Access

One of the most notable aspects of this release is the pricing change. OpenAI has cut the cost of API access for the GPT-6 Sol and Luna models by half compared to what developers were paying for the previous 5.6 series equivalents. The company attributes this reduction to advances in how the models handle caching and inference - technical improvements that allow the same level of output to be delivered at lower computational cost.

For developers and businesses building applications on top of OpenAI's infrastructure, this kind of price reduction can have a meaningful impact on the economics of deploying AI-powered features at scale. The move also signals that OpenAI is actively working to lower the barrier to entry for organizations that may have found earlier pricing prohibitive.

Improved Factual Accuracy and Fewer Coding Errors

Beyond pricing, OpenAI is also highlighting measurable gains in factual reliability. According to the company, GPT-6 Sol produces roughly half as many factual errors as its predecessor when evaluated against an internal benchmark. That benchmark is based on real-world conversations - with user identifiers removed - in which people flagged incorrect responses from OpenAI's models. The company says GPT-6 Sol now performs at a level of reliability comparable to GPT-6 Astra, but at a significantly lower price point.

"On our internal factuality evaluation, which is based on de-identified real-world conversations where users flagged mistakes by our models, GPT-6 Sol makes about half as many mistakes as its predecessor, reaching Astra-level reliability at much lower cost."

In addition to factual accuracy improvements, OpenAI notes that the new models also demonstrate a lower rate of errors in coding tasks - an important consideration given that Sol is specifically targeted at developers and technical users who rely on precise, functional output.

Competitive Positioning Against Anthropic

As has become a consistent pattern with OpenAI's model launches, the company used this announcement to draw direct comparisons with its primary rival, Anthropic. OpenAI claims that both GPT-6 Sol and Luna outperform Anthropic's leading models - including its Fable and Opus offerings - across a range of tasks.

The timing of the competing releases underscores just how fierce the competition between the two AI labs has become. Anthropic published an updated version of its Opus 5.5 model just 90 minutes before OpenAI made its GPT-6 Sol and Luna announcement public - a near-simultaneous launch that speaks to the rapid pace of development and the strategic importance each company places on being perceived as the industry leader.

Availability and Rollout Details

The updated Sol and Luna models are being made available across several of OpenAI's platforms. GPT-6 Sol and Luna are now accessible through ChatGPT Work and Codex for most users on paid subscription plans, as well as through the ChatGPT API for developers. Luna is also being made available through the ChatGPT desktop application and for users on the Free and Go tiers, broadening access to a much larger portion of OpenAI's user base.

OpenAI indicated that the models would be rolled out gradually across the ChatGPT application and website over the course of the day, meaning some users may gain access sooner than others depending on their account type and region.

The release continues OpenAI's pattern of rapidly iterating across its model portfolio, pushing improvements in both performance and cost efficiency as the competition among leading AI laboratories continues to intensify.