OpenAI has officially launched its GPT-5.6 series, introducing three new models: the flagship Sol, the balanced Terra, and the budget-friendly Luna. The release, which aims to address enterprise concerns regarding high operational costs, is available immediately across ChatGPT and the company’s API.
According to OpenAI, the new models prioritize “cost per task” improvements. Sol is priced at $5 per million input tokens, while Terra and Luna offer lower-cost alternatives at $2.50 and $1.00 respectively. To further support developers, OpenAI has updated its billing structure, including new prompt caching rules that reduce read costs to 10% of the standard rate.
A significant shift in this release is the integration of the standalone Codex app into ChatGPT, effectively creating a unified AI super app. The new architecture introduces “Ultra” and “Max” modes, allowing users to leverage multiple parallel agents—scaling up to 16—to improve reasoning and task execution speed. Early testing indicates strong performance in coding and web-browsing benchmarks, with the Sol model setting a new record on the Artificial Analysis Coding Agent Index.
While OpenAI claims competitive advantages in efficiency, the company’s benchmark data shows mixed results compared to Anthropic’s Claude Fable 5. While Sol leads in specific coding and operational tasks, Fable 5 retains higher scores in specialized areas like FrontierMath and SWE-Bench Pro. OpenAI also confirmed that the new models feature enhanced self-auditing capabilities, allowing the AI to inspect its own visual and functional outputs before completion.