OpenAI ships GPT-6 Sol and Luna at half the price with fewer factual errors

OpenAI has broadened its GPT-6 lineup with refreshed versions of its smaller Sol and Luna models, which the company says cost half as much to run as their predecessors while making fewer mistakes. The launch follows GPT-6 Astra, released earlier this month and positioned by OpenAI as its most capable model to date, according to TechCrunch.

Two tiers for two kinds of work

Sol and Luna, first introduced earlier this year, occupy different points in OpenAI's model hierarchy. Sol is built for demanding work such as coding, while Luna targets what OpenAI describes as clerical-style jobs: high-volume tasks with a well-defined outcome, such as summarising documents, pulling out specific information or answering quick questions.

In its announcement, OpenAI framed the new releases as an extension of Astra rather than a leap beyond it. The flagship "introduced a new generation of intelligence," the company said, and these models make that intelligence "more efficient and accessible."

Pricing cut in half

The headline change for developers is cost. TechCrunch reports that the GPT-6 series of Sol and Luna is available at half the price of the 5.6 series, a reduction OpenAI attributes to improvements in caching and inference. For teams running large volumes of API calls, a gap that wide can change which models are economical in production.

Fewer mistakes, by OpenAI's own benchmarks

OpenAI also claims accuracy gains. It says the new models produce more factually accurate answers and have a lower error rate on coding tasks. The company quantified the factuality claim using an internal evaluation built from de-identified real-world conversations in which users flagged mistakes by its models. On that benchmark, OpenAI said GPT-6 Sol makes roughly half as many errors as its predecessor and reaches "Astra-level reliability at much lower cost."

These are OpenAI's own figures. No independent benchmark results were included in the announcement, so the comparisons should be read as vendor claims.

A ninety-minute gap in the Anthropic race

As with previous launches, OpenAI asserted that the new Sol and Luna models handle tasks substantially better than the top models of its chief rival, Anthropic, naming Fable and Opus specifically. The competitive timing is hard to miss: according to TechCrunch, Anthropic shipped a new version of Opus 5.5 just ninety minutes before OpenAI's release, underscoring how closely the two companies are tracking each other's launch cycles.

Where the models are available

The updated Sol and Luna are available now in ChatGPT Work and Codex for most paid accounts, and through the ChatGPT API. Luna is also reaching the desktop app as well as Free and Go users. OpenAI said it expects the models to roll out gradually across the ChatGPT app and website throughout the day.

Why it matters

Two things make this release more than a routine mid-cycle refresh. First, halving the API price of Sol and Luna lowers the cost floor for production AI workloads, which matters to developers weighing whether smaller models can replace expensive flagships in pipelines that run at scale. Second, reliability — measured here as the rate at which users catch factual mistakes — remains one of the main barriers to deploying language models in clerical and coding workflows. If OpenAI's internal numbers hold up outside the lab, the claim that Sol now matches Astra-level reliability at a fraction of the cost changes the value equation for a wide range of applications. And with Anthropic countering within the hour, the pace of the model race shows no sign of slowing.