ChatGPT 5.6 Release: What's New in GPT-5.6

ChatGPT 5.6 Release: What Actually Changed

GPT-5.6 landed on July 9, 2026. It is better at coding, more reliable on long jobs, and less likely to lose the plot halfway through a complicated request.

This is not the kind of update where ChatGPT suddenly looks completely different. Most of the changes sit underneath the interface. You notice them when a task involves several files, a long list of constraints, or enough steps that an older model might drift off course.

There are also a few changes around the wider Codex experience, including optional desktop pets that show task status. Those are worth knowing about, but the main story is GPT-5.6 itself: a model designed to finish more of the job with less supervision.

First, Is It ChatGPT 5.6 or GPT-5.6?

The official name is GPT-5.6. ChatGPT is the app; GPT-5.6 is the new family of models inside it.

In regular ChatGPT, you will mostly see GPT-5.6 through the reasoning controls rather than a dramatic redesign. GPT-5.5 Instant is still there for quick questions. Choose Medium or High for something harder and ChatGPT uses GPT-5.6 Sol. On plans that include them, Extra High and Pro give the model more time to work.

So no, everything you type into ChatGPT is not suddenly running on the most expensive GPT-5.6 model. OpenAI is still matching different jobs with different amounts of compute.

The Three Models Without the Marketing Fog

GPT-5.6 comes in three versions: Sol, Terra, and Luna. The names sound grander than the decision really is.

Sol is the one you use when the task is genuinely difficult. It is built for long coding sessions, research, design work, cybersecurity, and jobs that involve several tools or files. This is the flagship model and the main upgrade paid ChatGPT users will notice.

Terra sits in the middle. It is cheaper than Sol and roughly aimed at the kind of everyday professional work that does not need the biggest model in the room. You cannot pick it in a normal ChatGPT conversation, but it is available in Codex, ChatGPT Work, and the API.

Luna is the fast, inexpensive option. It makes more sense for high-volume tasks, quick transformations, and smaller steps inside an automated workflow. Like Terra, it lives outside the standard ChatGPT model picker.

The practical version: use Sol when being wrong or getting stuck would waste your afternoon. Use Terra for the broad middle. Use Luna when speed and cost matter more than squeezing out the best possible answer.

Codex Also Has Optional Pets

Codex Pets are small animated companions in the desktop app. They show whether Codex is working, has finished, or needs input. Type /pet to turn one on and use /hatch to generate a custom pet.

This is a Codex interface feature, not part of the GPT-5.6 model. It launched on May 1, 2026, about two months before GPT-5.6, and does not change the model's quality or speed.

What GPT-5.6 Actually Feels Better At

The biggest improvement is not that GPT-5.6 knows one more fact than the previous model. It is that the model is harder to knock off course.

Give an older model a messy request with six constraints and it may complete the obvious part, forget two details, then confidently hand the cleanup back to you. GPT-5.6 is designed to keep the original goal in view for longer, use tools more sensibly, and check what happened before declaring victory.

That should show up most clearly in a few places.

Coding That Goes Beyond Producing a Snippet

GPT-5.6 is better suited to working inside an actual codebase: finding the relevant files, making changes, running commands, reading failures, and trying again. A good coding agent is not just a code generator. It needs to survive the boring middle of the job.

Codex also gets max and ultra reasoning options. Max gives the model more time to think. Ultra can split a hard job across subagents and bring the results back together. Those modes are useful for a migration or a stubborn production issue. They are overkill for renaming a button.

Frontend Work That Looks Less Generic

OpenAI made a point of highlighting design in this release, and this may be one of the more noticeable upgrades. GPT-5.6 can generate an interface, inspect the rendered page, spot obvious visual or functional problems, and refine it.

That matters because valid frontend code can still produce a bad interface. The interesting part is not another model that can write JSX. It is a model that can look at the page afterward and notice that the mobile layout breaks or the hierarchy feels wrong.

Presentations and Spreadsheets That Need Less Repair

GPT-5.6 is better at following an existing presentation template instead of inventing a new style on every slide. It can pay attention to layouts, typography, spacing, colors, and Slide Master rules. OpenAI says it also handles spreadsheet structure, equations, and financial models more carefully.

This sounds less exciting than a new chatbot demo, but it solves a common problem: AI can draft the content in minutes, then leave you with an hour of formatting. A useful model should reduce both parts of the job.

Longer Research and Computer Tasks

The new model is stronger at browsing, using software, and turning a pile of source material into a finished document. The promise is fewer moments where it stops at a summary when you asked for a decision, a report, or an editable deliverable.

Benchmarks support that story, but there is no need to memorize them. The real test is whether GPT-5.6 can take a task from the awkward beginning to a result you would actually share.

Who Can Use GPT-5.6?

The rollout started globally on July 9 and may take time to reach every account.

In standard ChatGPT chats, Plus users get GPT-5.6 Sol through Medium and High reasoning. Pro, Business, and Enterprise plans also include higher reasoning options, with Sol Pro reserved for the hardest jobs. Free and Go accounts do not get Sol in regular chats.

Codex is a little different. Free and Go users can access Terra, while paid plans can choose among Sol, Terra, and Luna. If GPT-5.6 is missing in Codex, check that the app or CLI is current because the new models require a recent version.

What Does the API Cost?

OpenAI charges per one million tokens:

• Sol: $5 input, $30 output
• Terra: $2.50 input, $15 output
• Luna: $1 input, $6 output

Those prices make the three-model setup easier to understand. Sol is expensive enough that developers will not want it handling every tiny step. Terra and Luna exist so an app can save the heavy reasoning for moments that deserve it.

The API also adds Programmatic Tool Calling, which lets GPT-5.6 coordinate tools and process intermediate results with code running in memory. Multi-agent support is entering beta as well. Both features are aimed at making agents less chatty internally and more capable of completing a full workflow.

Is the Upgrade Worth Caring About?

If you use ChatGPT for short rewrites and casual questions, GPT-5.6 may not change much. Instant is still faster, and the new reasoning modes can take longer.

If you use AI to build software, research a topic, produce a deck, or work across several files, the upgrade is more meaningful. GPT-5.6 is trying to reduce the amount of supervision between a prompt and a finished result. That is a less dramatic promise than "an AI that knows everything," but it is much closer to what people need at work.

Overall, GPT-5.6 matters less as a new name in the model picker and more as an attempt to make longer AI-assisted work feel dependable.

Quick Questions

When did GPT-5.6 launch?

OpenAI started the broad rollout on July 9, 2026, after a limited preview in June.

Are Codex Pets part of GPT-5.6?

No. Codex Pets launched on May 1, 2026. They are an interface feature in the Codex desktop app, while GPT-5.6 is the model doing the underlying work.

How do I get a Codex pet?

Type /pet in Codex. Use /hatch if you want to generate a custom one.

Is GPT-5.6 free in ChatGPT?

GPT-5.6 Sol is not available to Free or Go users in standard chats. Those users can access GPT-5.6 Terra through Codex.

Does GPT-5.6 replace GPT-5.5 Instant?

No. Instant remains the default for fast everyday questions. GPT-5.6 Sol powers the deeper reasoning options on eligible plans.

Sources

OpenAI: GPT-5.6 release announcement

OpenAI Help Center: GPT-5.6 in ChatGPT

Codex Pets announcement coverage