OpenAI releases Sol flagship model with 91.9% Terminal Bench score, adds Terra and Luna free tiers

OpenAI released three models on July 9: Sol, Terra, and Luna. Sol, positioned above the existing Mythos tier, will roll out to paid users within 24 hours. The company also announced ChatGPT Work, a new desktop application, and a hosted sites feature for deploying websites directly through ChatGPT.

Sol is OpenAI's new flagship reasoning model. Early testers report it scores 91.9 percent on Terminal Bench 2.1 Ultra, edging past prior leaderboard entries in that configuration. Testers highlight Sol's UI assembly capability—the model builds functional interfaces for apps and websites rather than generating raw code that requires manual wiring. That distinction matters for practitioners who need production-ready components, not boilerplate. OpenAI has not disclosed parameter counts, context lengths, or training data details for Sol or the other two models.

Terra matches Fable 5 performance and will be available to free-tier users. Luna is also free. The tiered release—one paid flagship, two free alternatives—mirrors the company's existing model lineup, where capability gaps between tiers determine access.

ChatGPT Work targets enterprise users, though pricing and feature details remain undisclosed. The new desktop application is now available across platforms. The hosted sites feature lets users deploy websites directly through ChatGPT without separate hosting infrastructure—practitioners can generate a site in the interface and publish it immediately, compressing the workflow for prototyping and small-scale deployments, though production requirements like custom domains, SSL, and backend integration remain unclear.