OpenAI released GPT-6 Sol and GPT-6 Luna today, cutting API pricing to half the cost of the outgoing 5.6 series and claiming Sol makes about half as many mistakes as its predecessor. The launch extends the GPT-6 generation OpenAI opened earlier this month with Astra, its flagship. Anthropic pushed a new Opus 5.5 roughly 90 minutes before the OpenAI announcement, a timing choice that leaves little doubt about how tightly the two labs are tracking each other's release calendars.
Sol is positioned for complex tasks like coding, while Luna is pitched at high-volume clerical work such as summarizing documents, extracting information, and answering short questions. Both were introduced in earlier form this year and now step up to the GPT-6 architecture that underpins Astra.
“GPT-6 Astra introduced a new generation of intelligence; these models extend its benefits by making that intelligence more efficient and accessible”— OpenAI, company statement
The pricing move is the sharpest signal in the release. Cutting API costs in half against the 5.6 tier — attributed by OpenAI to caching and inference improvements rather than a smaller model — puts downward pressure on the entire mid-tier market, where developers pick models on a cost-per-token basis and switch when the math shifts. Anthropic's Fable and Opus lines sit in the same competitive band, and OpenAI is explicit throughout the announcement that Sol and Luna are meant to outperform them.
Key facts
- 01GPT-6 Sol and Luna are priced at half the cost of the 5.6 series, with OpenAI crediting caching and inference gains.
- 02Sol produces roughly half as many mistakes as its predecessor on OpenAI's internal factuality evaluation.
- 03Anthropic released Opus 5.5 just 90 minutes before the OpenAI launch.
- 04Sol targets complex coding work; Luna is aimed at high-volume clerical tasks like summarization and extraction.
- 05The models are live in ChatGPT Work, Codex, and the ChatGPT API, with Luna reaching Free and Go users in the desktop app.
Factuality is the second lever. OpenAI's internal evaluation draws from de-identified real-world conversations in which users flagged model mistakes, and the company says Sol now approaches Astra-level reliability at a fraction of the cost. Coding error rates are also down, which matters most in Codex, where a wrong suggestion doesn't just annoy a user — it ships into a codebase.
Distribution is broad from day one. Sol and Luna are available in ChatGPT Work and Codex for most paid accounts and through the ChatGPT API. Luna is also going into the desktop app and reaches Free and Go users, with OpenAI rolling the models into the main ChatGPT app and website gradually through the day.
“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”— OpenAI, launch announcement
The Astra release earlier this month was pitched by OpenAI as the world's best model for computer work and coding. Sol and Luna extend that generation downward on the price-performance curve, giving OpenAI a full stack — flagship at the top, cost-optimized workhorses below — to match how enterprise buyers actually assemble their model portfolios. Most production deployments route the majority of traffic to smaller models and reserve the flagship for the hardest calls.
The Anthropic response, an Opus 5.5 push landing 90 minutes ahead of OpenAI, underscores how narrow the release windows have become. Both labs are now shipping incremental point releases against each other on the same day, with pricing and factuality claims doing most of the differentiation work rather than raw capability jumps. The frontier moves less in leaps and more in a steady grind of cheaper tokens and fewer errors.
One thing the announcement doesn't detail is the specific benchmark deltas against Anthropic's Fable and Opus lines beyond the general claim of substantial improvement. OpenAI's factuality number is measured against its own prior model on its own internal evaluation, not a third-party benchmark, so cross-lab comparisons will have to wait for independent testing.
The competitive read is that OpenAI is using price to defend the middle of the market, where volume lives. Halving API costs on Sol and Luna forces every developer currently paying Anthropic rates for Fable or a mid-tier Opus to run the math again, and forces Anthropic to answer on price or on a capability lead large enough to justify the gap. With Opus 5.5 arriving the same afternoon, that answer is already in motion — but the pricing floor just dropped, and mid-tier model economics for the rest of 2026 will be written from here.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
