OpenAI is rounding out its GPT-6 family. Weeks after launching GPT-6 Astra — the flagship model it called its most capable yet — the company has released updated versions of Sol and Luna, the smaller, cheaper tiers that sit beneath Astra in its model hierarchy. The pitch, in OpenAI’s own words, is that if Astra introduced a new generation of intelligence, these models are meant to make that same intelligence more efficient and more accessible.

What Sol and Luna are actually for

The Sol and Luna names aren’t new — OpenAI introduced the first versions of both back in July as part of its GPT-5.6 lineup, and this release essentially upgrades them to the GPT-6 generation rather than introducing a new concept. The division of labor between the two remains the same: Sol is positioned for complex tasks like coding, while Luna is built for high-volume, clearly defined work — summarizing documents, extracting information from text, or answering quick factual questions.

The headline change: half the price

The most concrete upgrade here isn’t a new capability — it’s cost. OpenAI says the GPT-6 versions of Sol and Luna are available via its API at half the price of the GPT-5.6 versions, a reduction the company attributes to improvements in caching and inference efficiency rather than any change in underlying compute. For businesses and developers running high-volume workloads through Sol or Luna, that’s a direct, immediate cost reduction rather than a promised future capability.

Fewer mistakes, OpenAI says

Beyond pricing, OpenAI is emphasizing accuracy gains. The company says GPT-6 Sol makes roughly half as many factual errors as its predecessor, based on an internal evaluation built from de-identified real-world conversations where users had flagged model mistakes — and claims Sol now reaches reliability levels comparable to the flagship Astra model, but at a fraction of Astra’s cost. OpenAI also says both new models show a lower error rate specifically on coding tasks.

These are OpenAI’s own internal benchmarks rather than independently verified third-party results, so they’re worth treating as the company’s claim about its own product rather than a neutral comparison — a caveat that applies to essentially every AI lab’s self-reported model benchmarks, not uniquely to OpenAI.

A pointed jab at Anthropic

As has become standard with OpenAI’s recent model launches, the announcement doesn’t just compare Sol and Luna to their own predecessors — it directly targets rival Anthropic, with OpenAI repeatedly claiming its new models outperform Anthropic’s Fable and Opus models on a range of tasks. The timing made the rivalry particularly visible this time: Anthropic released an updated version of Opus 5.5, with lower prices and performance it says matches its own Fable model, just 90 minutes before OpenAI’s Sol and Luna announcement — a scheduling coincidence, or a deliberate one, that underscores how tightly the two labs are now racing each other on both capability and price.

Rollout details

The updated Sol and Luna models are now live in ChatGPT Work and Codex for most paid accounts, and available through OpenAI’s API. Luna is additionally rolling out in the ChatGPT desktop app and for Free and Go tier users. OpenAI says the broader rollout to the ChatGPT app and website will happen gradually over the course of the day.

Why it matters

Taken together with Astra’s launch earlier this month, OpenAI now has a fully refreshed three-tier GPT-6 lineup — a flagship model for maximum capability, and two lighter, cheaper models aimed at cost-sensitive, high-volume use cases. That structure mirrors a broader industry pattern in 2026, where AI labs are no longer competing purely on raw model capability at the top end, but increasingly on how cheaply and reliably they can serve everyday, high-volume tasks — the kind of work that makes up the bulk of actual enterprise AI spending, even if it generates far less headline attention than flagship model releases.