Strike While the Iron Is Hot! All Three GPT-5.6 Models Revealed—Set for July 7?

GPT-5.6, to be released next week?

Right yesterday, netizens were excited to discover: three major sub-model identifiers—GPT-5.6 Sol, Terra, and Luna—appeared in the underlying code of the Codex app.

Even more exciting is that a brand-new “speed dial” feature also shows up in the code.

This suggests that users can freely adjust between speed and quality according to their needs, which will undoubtedly bring everyone an unprecedented level of control experience.

According to the leak, OpenAI has already set a strict internal directive: the target window for the release of GPT-5.6 is aimed directly at next Tuesday (July 7) through July 9.

Why July 7? Because that day happens to be the gap period when the specific quota scheme for Claude Fable 5 becomes invalid.

This is a business siege precisely calculated down to the hour.

Recently, because of a series of questionable moves, Anthropic has driven countless developers crazy, forcing Google Gemini 3.5 Pro into an urgent “rebuild.” And OpenAI is taking advantage of this moment to make its move and scoop up users at a bargain!

Breaking down the Codex code: Sol, Terra, and Luna are coming

“Honestly, OpenAI is acting like nothing happened—quietly stuffing model names into dead code, like we wouldn’t notice,” one netizen joked.

Since the limited release of GPT-5.6, geeks have been closely watching every front-end update from OpenAI.

Finally, in a recent merge of the Codex app, someone discovered traces of GPT-5.6.

And netizens also shared short videos. Although they currently can’t successfully call the model due to backend API limitations, in the front-end pop-up window the appearance of the three models and the brand-new “speed selector” are already clearly visible.

In addition, the code also faintly shows the words “Sol Ultra.” Industry speculation is that Sol Ultra will be a “wild card” directly targeting the competitors’ top-tier flagship—its performance should be on par with Fable 5, but its price will be far more friendly.

Besides these three models, the code also reveals a key piece of information: the much-anticipated “real-time voice support” is still under development, and is likely not to go live directly next week.

Leaked real-world tests leak out: GPT-5.6 vs Fable 5

Although most people haven’t gotten to use it yet, a few players who received internal test access have already shared comparative evaluations of GPT-5.6 in actual engineering environments.

The result is four words—dimensionality reduction.

Round 1: A tight contest at the limits of efficiency and understanding

Overseas tech blogger Shivam shared his experience using GPT-5.6-terra and Fable 5 to solve the same complex technical prompt.

Fable-5 begins with the 100% quota limit of a 5-hour session. That model wildly “thinks” in the background, burning 21% of the quota limit before finally producing a response that, to his surprise, consisted of asking a bunch of cross questions—making him re-confirm the technical details he wanted to solve.

For the same task, GPT-5.6-terra consumed only 13% of the quota, with an astonishing response speed.

It didn’t waste words. Instead, it directly and efficiently listed several different methods and architectural paths to solve the problem, then quickly began executing.

Shivam was blunt: When using Fable, I kept wondering whether it would suddenly downgrade to Opus 4.8; GPT-5.6-terra’s decisiveness made me feel extremely comfortable.

Round 2: Hardcore WebGL frontend mini-game “blind test”

Oracle director Gilson Melo gave GPT-5.6 High and Fable 5 High a hardcore challenge:

In a single HTML file, build from scratch a fully functional browser-based game (using WebGL or HTML5 Canvas). The game must include real-time rigid body physics, gravity, and a user-controlled paddle/spaceship mechanism. Write the complete CSS, JS, and HTML—do not omit any logic. It must support mouse dragging with real-time responsive physics feedback.

This challenge really tests the model’s ability to handle extreme details, generate long code without shrinking, and calculate underlying physics formulas.

The two models showed different strategies in their workflows.

Fable 5 High performed impressively: it confidently generated the entire game’s full code in one go.

By contrast, GPT-5.6 High paused twice in a very human-like way during generation, and proactively asked the developer to clarify two final key decisions.

Even more impressively, without being asked, it took the initiative to add sound effects to the game.

The final results showed that GPT-5.6 High won a more solid score in overall game experience, the smoothness of physics collisions, and the robustness of details.

In short, both testers believed that GPT-5.6 is one step ahead in efficiency and response style—especially in clarity and speed when handling complex tasks.

From this result, it’s definitely necessary to look forward to next week’s GPT-5.6 rollout.

Precisely timing it: OpenAI strikes while the fire is hot to steal users

If the model leak was an accident, then the release timing is definitely a carefully planned arrangement.

OpenAI plans to release GPT-5.6 on July 7, right on the day Claude users lose access to Fable 5.

Recently, Claude has lost a significant number of users, and OpenAI is eyeing the opportunity to take them all over.

An insider revealed: “GPT-5.6’s usage quota limit will be significantly loosened, more generous than Fable 5. Stricter safety guardrails are also being rolled out gradually, but they won’t be as aggressive as Fable’s in ways that affect normal usage.”

User dissatisfaction is surging—OpenAI is right on time to poach people

Compared with that, Anthropic has been facing a lot of public complaints lately.

Although Claude Fable 5 has just returned, it has already sparked strong user dissatisfaction.

Ask a few random questions, and Fable 5 will downgrade to Opus 4.8.

Biomedical engineer Derya Unutmaz tried to get Fable 5 to explain the word “human.”

After only typing “Explain human,” the model thought for a few seconds and then popped up a card saying “Switched to Opus 4.8,” because Fable 5’s safety mechanism determined that there was content in that message that needed to be intercepted.

Even funnier is that semiconductor analyst Dylan Patel asked an extremely simple question: “How many letters ‘r’ are in the word ‘raspberry’?”

This question was also intercepted, and the interface popped up “Chat paused,” indicating that Fable 5’s safety mechanism will intercept most topics related to cybersecurity or biology.

Also, the hallucination issues of Opus 4.8 recently have been very serious—so serious that other people’s information even appears in its own conversations.

This cliff-like drop in user experience is exactly what creates the best window for OpenAI to poach users.

And what’s more, GPT-5.6 is also likely to offer a cost advantage.

The leak shows that GPT-5.6 Sol will be more than twice as cheap as Fable 5 because it has higher token efficiency. But the key question is whether its performance is enough to be comparable to Fable 5.

Some predict that Sol Ultra should be able to match Fable 5 while also being cheaper. If that prediction comes true, OpenAI will clearly win over the competition in cost-effectiveness.

Developers reminded: 4 Codex quota resets—don’t let them go to waste

Finally, here’s a “take advantage / avoid pitfalls” guide to sync up with all hardcore developers who are preparing to return to Codex.

According to Reflection CTO’s deep dive, if you previously stored up 4 rate-limit reset credits in Codex, immediately check your account back end.

OpenAI’s official underlying rules show that these reset credits are only valid for 30 days. If your first credit arrived around June 11 or 12, then around July 12 they will begin expiring in batches!

If you want to know your exact expiration time, you can have Codex call your ChatGPT token to request this back-end API: GET https://chatgpt.com/backend-api/wham/rate-limit-reset-credits.

You will receive a JSON response similar to the following:

If GPT-5.6 really is unlocked on time next Tuesday, you’ll only have a short window of 4 to 5 days to spend your first reset credits.

Next Tuesday, OpenAI will most likely send everyone another brand-new Reset. So these days, hurry up and use your old credits where it matters.

GPT-5.6, see you next week!

Source: New Zhixing

Risk Warning and Disclaimer

        The market is risky; invest with caution. This article does not constitute personal investment advice, and it does not take into account any individual users’ specific investment goals, financial conditions, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article align with their specific circumstances. You take responsibility for any investment made based on this.