Why Codex Disappeared: GPT‑5.6 Full Release Merges Codex into ChatGPT

When the author opened the computer this morning, the familiar Codex app was gone; OpenAI had merged Codex into the new ChatGPT desktop application as part of the GPT‑5.6 full release announced on July 9.

GPT‑5.6 arrives in three versions—Sol, Terra, and Luna—each with its own API pricing per million tokens. Sol costs $5 input / $30 output and targets flagship, hard‑problem workloads; Terra costs $2.5 input / $15 output for balanced use; Luna costs $1 input / $6 output for lightweight, high‑frequency tasks.

OpenAI states that every token now delivers more intelligence and higher dollar performance. Compared with Claude Fable 5, Sol’s input price is half and its output price 40 % lower; prompt‑caching adds a 90 % discount, further reducing long‑context costs.

Sol can generate up to 750 tokens per second (≈500‑700 Chinese characters), dramatically shortening the “model thinking” wait time. In the Artificial Analysis benchmark, GPT‑5.6 consumes noticeably fewer tokens, especially for inference, yielding high unit efficiency.

Agents' Last Exam scores: Sol 53.6 (13.1 points above Fable 5) while achieving roughly one‑quarter the cost.

Terra and Luna claim about 1/16 the cost for certain tasks.

Early community tests confirm that although Sol may not dominate the hardest challenges, its cost‑performance makes it better suited for everyday office work.

ChatGPT Work, highlighted by Sam Altman as a “really big deal,” is positioned as the primary product built on the GPT‑5.6 model base.

The integration splits Codex into two layers: the original Codex entry remains for developers, handling code, repositories, PRs, and multi‑repo engineering tasks; the Agent capabilities are abstracted into ChatGPT Work, delivering a pipeline of “accept task → read context → call tool → execute stepwise → deliver result” for knowledge workers.

ChatGPT now offers three clear entry points:

Chat : conversational Q&A.

Codex : coding and engineering tasks.

Work : general office agents that operate across applications and files, capable of assisting a project for hours.

The significance is threefold: (1) AI competition shifts from “who has the smartest model” to “who can complete more tasks at lower cost,” with GPT‑5.6’s pricing, speed, and token efficiency turning intelligence into a cheap, consumable resource; (2) coding agents are becoming production‑grade agents, delivering Codex’s capabilities to non‑technical users via ChatGPT Work; (3) the ecosystem battle is underway, as GPT‑5.6 becomes the default model for Microsoft 365 Copilot while rivals like Claude adjust limits, indicating that the second half of 2026 will be fought over everyday workflow dominance.

In essence, ChatGPT is no longer just a chat tool but a complete work platform that bundles a model foundation, engineering engine, and universal executor, marking the start of a true AI‑powered office era.