OpenAI Unveils GPT-5.6 with Sol, Terra, Luna Tiers and Codex Merged into ChatGPT

On July 9, OpenAI released GPT‑5.6, announcing a new naming convention where the numeric part indicates the generation (5.6) and the celestial name denotes the capability tier. Three tiers are offered:

Sol (Sun) : flagship model, optimized for inference and programming; $5 per million input tokens and $30 per million output tokens. It achieved 80 points on a programming‑agent benchmark, surpassing all competitors and, compared with Fable 5, delivering the same score with half the token count and latency, reducing cost by roughly one‑third.

Terra (Earth) : balanced model with half the price of Sol, $2.5 per million input tokens and $15 per million output tokens.

Luna (Moon) : lightweight, cheapest tier at $1 per million input tokens and $6 per million output tokens. Notably, Luna’s post‑training was performed by Sol itself, an example of “AI training AI”.

The image below illustrates the tier hierarchy.

In parallel, the standalone Codex app was discontinued and its functionality merged into the new desktop version of ChatGPT; the previous version was renamed “ChatGPT Classic”. The Atlas browser was also retired, with its features moved to a Chrome sidebar extension.

OpenAI reports that Codex now has over 5 million weekly active users, more than 1 million of whom use it for tasks unrelated to coding, indicating that the tool has become a general‑purpose assistant. Consequently, OpenAI integrated Codex into ChatGPT, creating “ChatGPT Work”. This agent can operate across applications and files, interact with Slack, Teams, Google Drive, and email, run for hours, and output documents, spreadsheets, presentations, or shareable web pages.

An “ultra” mode enables multiple sub‑agents to collaborate autonomously, handling task decomposition and division of labor. In the Terminal‑Bench evaluation, this mode achieved a 91.9 % score, the highest reported.

The model’s design capabilities have also improved; it can render pages, check visual quality, and only deliver finished results. An evaluation partner gave the new design quality a score of 4.4, up from 4.0 for GPT‑5.5.

Overall, the merger of Codex reflects a broader trend of consolidating separate AI tools—coding, writing, design—into a single conversational interface. OpenAI has taken the first step; the article speculates which tool might be integrated next.