GPT-6 Sol and Luna ship in the API, 50% cheaper, on DevDay eve
OpenAI halved both models' rates, added 90%-off cache reads, and put Luna in the free desktop app — plus a font compiler that renders your tokens.
Copy markdown
GPT-6 Sol and Luna are live today
OpenAI dropped both models into the API as `gpt-6-sol` and `gpt-6-luna`, rolling out through the day — and a day ahead of its DevDay keynote. Sol makes about half as many mistakes as GPT-5.6 Sol, keeps the strong coding, and both write with less jargon.
Both cost 50% less, and cached reads are 90% off
Sol is $2/$10 per million input/output tokens, Luna just $0.10/$0.50 — half the old GPT-5.6 rates. Cached input reads now get a 90% discount by default, and you can change reasoning effort or tools without busting the cache. A new caching dashboard and diagnostics tool track your hit rates.
Free and Go users get Luna in the desktop app
Luna lands for Free and Go tiers in the ChatGPT desktop app, while Plus, Pro, Business, Enterprise and Edu get both models in ChatGPT Work and Codex. If you code on a paid plan, your default just got cheaper and more accurate.
A font compiler that makes your tokens visible
A new browser tool fuses any font with a tokenizer (Claude, OpenAI, DeepSeek, Gemini) so every LLM token renders as an equal-width or color-coded glyph, compiling a TTF entirely client-side. Drop it into your editor, Discord or Slack to see exactly why a prompt costs what it does.