GPT-5.6 Released July 9: How to Use ChatGPT, Codex, API and Why It May Be Hidden

Rollout timing and layered release

OpenAI released a limited preview of GPT‑5.6 on 26 June 2026. Media reports on 8 July 2026 indicated that a broader rollout began on 9 July 2026. The rollout is staged, not a simultaneous full‑scale launch for all accounts.

New flagship models are distributed gradually based on product entry point (ChatGPT web, ChatGPT mobile app, Codex, OpenAI API), subscription tier (Pro, Plus, Team, Enterprise), geographic region, and OpenAI’s risk‑control policies. Consequently, some users see the model in ChatGPT, others only in Codex or the API, and a higher‑tier subscription does not guarantee immediate visibility.

GPT‑5.6 model family

GPT‑5.6 Sol – Flagship model. Targeted at complex programming, research analysis, network‑security defense, and long‑task agents.

GPT‑5.6 Terra – Balanced model. Suited for everyday office work, content writing, document analysis, standard code tasks, and enterprise assistants.

GPT‑5.6 Luna – Fast, low‑cost model. Designed for summarization, classification, customer‑service replies, lightweight Q&A, and batch automation.

In short, Sol offers the highest capability at the highest cost, Terra provides a cost‑effective balance for most production workloads, and Luna delivers cheap, high‑throughput handling of lightweight tasks.

How to access GPT‑5.6

1. ChatGPT (web)

If the model has been rolled out to your account, it appears in the model selector.

Open the ChatGPT website.

Log in to your account.

Open the model selector.

Look for options named GPT‑5.6, GPT‑5.6 Sol, GPT‑5.6 Terra or GPT‑5.6 Luna.

If none appear, wait for the next rollout phase; do not purchase “GPT‑5.6 access”.

Mobile‑app users should update the app, log out and back in, switch to the web version, and wait for server sync.

2. Codex

Codex is the primary entry for GPT‑5.6 Sol. OpenAI notes that Sol achieves a new best on the Terminal‑Bench 2.1 benchmark, which mimics real‑world command‑line engineering tasks.

Check whether GPT‑5.6 Sol appears in the Codex model list.

Verify that the Codex CLI offers a new model option.

Ensure IDE plugins are up‑to‑date.

Enterprise or team workspaces may receive priority access.

Look for higher inference intensity or fast‑mode options.

3. OpenAI API

Developers should inspect the OpenAI API console for the three model names and any prompt‑caching settings. Route requests by task difficulty:

Complex code, long‑running agents → use GPT‑5.6 Sol.

Daily workflows, document analysis → use GPT‑5.6 Terra.

Summarization, classification, lightweight automation → use GPT‑5.6 Luna.

This avoids sending every request to the most expensive model.

Plus / Pro users and visibility

Plus/Pro users are more likely to receive early access, but rollout may prioritize Pro, Team, Enterprise, Codex, or API users before Plus. The model may appear first for internal partners, then for Pro, then for Plus, and finally for free users.

Why some users still cannot see GPT‑5.6 after 9 July

Staged gray‑scale: internal partners → trusted collaborators → Pro/Team/Enterprise → Plus → free users.

Different entry points: the model may appear in Codex or the API before the ChatGPT web UI, or vice‑versa.

Account plan limitations: Sol has stricter access controls.

Regional or security restrictions: abnormal login patterns, unstable subscription status, or risky network environments can block access.

Enhanced safety policies: requests in dual‑use domains (e.g., cybersecurity, biology) may be rejected, slowed, or require extra checks.

Choosing the appropriate GPT‑5.6 variant

Sol – when to use

Complex code refactoring

Long‑duration debugging

Multi‑file project modifications

Advanced research and data analysis

Network‑security defense testing

Agent tasks with long execution chains

Terra – when to use

Everyday writing and documentation

Document summarization

Table analysis

Standard code Q&A

Enterprise knowledge bases

Meeting minutes and proposal drafting

High‑quality but not extreme workloads

Luna – when to use

Batch summarization

Content classification

Customer‑service replies

Tag generation

Simple Q&A

Data cleaning

Lightweight automation

GPT‑5.6 API pricing (per 1 M tokens)

GPT‑5.6 Sol – Input $5, Output $30

GPT‑5.6 Terra – Input $2.50, Output $15

GPT‑5.6 Luna – Input $1, Output $6

Prompt caching improvements

Explicit cache checkpoints are supported.

Minimum cache lifetime is 30 minutes.

Cache writes are billed at 1.25× the uncached input price.

Cache reads receive a 90 % discount on input pricing.

For enterprise knowledge bases, code‑base Q&A, long‑context agents, or multi‑turn tool calls, prompt caching can yield significant cost savings.

Implications for users in China

Ensure the ChatGPT account can log in normally.

Verify that the Plus/Pro subscription is active.

Confirm that GPT‑5.5 or the current latest model works without errors.

Check that file upload, browsing, image, and advanced research features are functional.

Make sure Codex or API endpoints are reachable.

Avoid purchasing “independent GPT‑5.6 permissions” from third‑party sellers; the reliable path remains official ChatGPT subscription, Codex, or the OpenAI API.