OpenAI Launches GPT-6 Sol and Luna at Half the Price
Sol handles the complex work and Luna the high-volume tasks: OpenAI's two new GPT-6 models cost half as much in the API as the 5.6 series.
In short
OpenAI shipped two more GPT-6 models on September 22, 2026 — Sol for complex work including coding and Luna for fast, high-volume tasks — and TechCrunch reports both cost half of the matching 5.6 series models in the API.
At a glance
- OpenAI released two GPT-6 models on September 22, 2026: Sol for complex work, Luna for fast high-volume tasks.
- TechCrunch reports API pricing at half the 5.6 series, which OpenAI credits to better caching and faster inference.
- Sol reportedly makes about half as many factual mistakes as its predecessor; no measured values were published.
- GitHub enabled Luna on five Copilot plans and Sol on four, with both reaching ten Copilot surfaces.
- Anthropic shipped Opus 5.5 roughly 90 minutes before OpenAI's announcement.
Two more members of the GPT-6 family arrived on September 22, 2026. GPT-6 Sol targets complex, multi-step work including coding; GPT-6 Luna targets clerical volume. TechCrunch reports that API access to the 6 series costs half what the equivalent 5.6 models did. The generation opened earlier this month with Astra.
Two models, two jobs
Luna is scoped to work with an obvious finish line — summarizing a document, pulling a field out of a form, answering a quick question. Sol is the one you point at a task that branches.
GitHub's own changelog frames Sol as a balanced choice for interactive and agentic coding, and Luna as the lowest-cost option in the GPT-6 lineup.
Price is the headline, not capability
OpenAI attributes the cut to caching and inference improvements rather than a smaller model, and claims Sol reaches Astra-level reliability at much lower cost. That reframes the buying question from which model is strongest to what a finished task costs.
One caveat on the numbers: OpenAI's announcement page returned HTTP 403 when we retrieved it on September 24, 2026. The half-price ratio is what the reporting supports; per-million-token figures are not something we were able to confirm.
Fewer mistakes, unpublished numbers
Sol is described as making about half as many mistakes as its predecessor on factuality evaluations, with lower error rates on coding too. The underlying scores are not in the reporting, and we could not check them against OpenAI's own documentation.
OpenAI further claims both models substantially outperform Anthropic's Fable and Opus. Treat that as a vendor claim until someone outside the company measures it.
Where you can already run them
Sol and Luna are live in ChatGPT Work, Codex and the API, with Luna also reaching the desktop app and Free and Go users. The rollout was staged across September 22, 2026.
In GitHub Copilot, Luna covers five plans — Pro, Pro+, Max, Business and Enterprise — while Sol covers four, skipping Pro. Both reach ten surfaces, from Visual Studio Code and JetBrains through Xcode and Eclipse. Billing is usage-based, and new models switch on by default for Business and Enterprise until an administrator disables them in model policy.
The 90-minute gap
Anthropic put Opus 5.5 out roughly 90 minutes ahead of OpenAI's announcement. The release calendar is now close enough that teams comparing the two will be comparing invoices as much as benchmarks.
FAQ
How much do GPT-6 Sol and Luna cost in the API?
TechCrunch reports half the price of the equivalent 5.6 series models. We could not verify absolute per-million-token pricing, because OpenAI's announcement page was unreachable when we fetched it.
What is the difference between GPT-6 Sol and GPT-6 Luna?
Sol is built for complex, multi-step tasks and coding. Luna is the lightweight model for high-volume work with a clear goal, such as summarizing documents or extracting specific information.
Are GPT-6 Sol and Luna available in GitHub Copilot?
Yes, since September 22, 2026. Luna is enabled on five Copilot plans and Sol on four, across ten surfaces from Visual Studio Code to Eclipse, billed on a usage-based model.
