OpenAI Releases GPT-6 Sol and Luna: 50% Cheaper API Pricing and Benchmarks

OpenAI has filled out its GPT-6 lineup with two new models: GPT-6 Sol and GPT-6 Luna. Both sit below GPT-6 Astra, which arrived earlier this month. The pitch is simple enough. Keep Astra for the hardest jobs, then use Sol and Luna when speed and cost matter more than raw firepower.

Both models are already live in the OpenAI API under the names gpt-6-sol and gpt-6-luna. They’re API-only, so there’s no option to download weights and run them yourself. OpenAI says they were trained with methods similar to Astra’s, and that the point was to carry Astra’s gains into models that are faster and cheaper to serve.

The pricing change is the headline here. Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20. Luna drops to $0.10 on input and $0.50 on output, from $0.20 and $1.20. OpenAI calls that a 50% cut against its GPT-5.6 promotional pricing, though Luna’s output price falls by a bit more than that.

OpenAI’s own benchmarks try to show why the split matters. Sol is aimed at complex coding and professional work, while Luna is pitched at high-volume everyday use. On AutomationBench 1.0.6, Sol at xhigh effort scores 33.2%, and OpenAI says that beats Claude Opus 5’s 26.9% at 11.1 times the cost. On Agents’ Last Exam, Sol at max effort scores 56.4%, ahead of Claude Opus 5 at 60% lower cost per task.

Coding and computer use are where the cheaper models get more interesting. On DeepSWE v1.1, Sol at max effort scores 68.8%, just 1.1 points behind Claude Fable 5 at xhigh, while Luna hits 66.6% and lands in the same neighborhood as Opus 5 and Fable 5 at medium effort. On OSWorld 2.0 offline, Sol at xhigh scores 60.5% versus 60.3% for Opus 5 at medium. OpenAI also says Sol makes about half as many mistakes as its predecessor on an internal factuality test using de-identified ChatGPT conversations where users flagged errors.

There’s a quieter but useful update buried in the release too: prompt caching. GPT-6 adds a caching system with higher hit rates by default, discounts of up to 90% on cached input reads, and new controls like a dashboard, diagnostics for cache misses, explicit breakpoints, and prewarming. OpenAI says GitHub has already seen more than a 50% cut in fresh prompt tokens needing processing. Availability is rolling out to the API and to ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, while Luna is also coming to the ChatGPT desktop app for Free and Go users. It is not yet in Chat.