Anthropic Fable 5.2 Leaked, Outperforms GPT-6 Astra in Tests

Just now, Fable 5.2 has also been revealed!

Many netizens have noticed that Fable 5.2 has entered gray testing.

A large number of real test cases have now appeared online.

Based on extensive overnight testing by users, this update is so substantial that Anthropic skipped the "5.1" designation entirely.

Under maximum reasoning mode, Fable 5.2's actual performance surpasses GPT-6 Astra!

Moreover, Opus 5.2 has also been announced—it may be released soon, possibly even today.

The brand-new Opus 5.2 model (internal code name: Opus-Next) also experienced a "launch-unplug-relaunch" scenario yesterday.

This week, none of the big tech companies in Silicon Valley released models—they were all testing in secret.

Looking at this chart, it's truly impressive.

Yesterday, Anthropic launched the gray test of Fable 5.2 in Claude Code, Chat, and Cowork.

Some requests that originally called Fable 5.1 are quietly redirected to Fable 5.2 in the background.

Now, more and more real-world test cases are being leaked online. Some people say: OpenAI is in trouble.

Case 1: High Reasoning Mode Delivers Higher Output Quality Than Astra!

Renowned AI testing blogger @notjazii discovered that their Fable 5.1 was routed to the new version and began a series of tests.

In high reasoning mode, he used the exact same prompt to compare Fable 5.2 and GPT-6 Astra.

The conclusion is: "This is a massive leap forward compared to the current Fable version. The output from Fable 5.2 is clearly superior to Astra!"

He tested the output under various modes and compared it with Fable 5.2 (suspected).

However, he also pointed out the cost under the physical laws of computing power—slower and more expensive.

To break through the ceiling in logical coherence and long-form reasoning, Fable 5.2 has adopted deeper "slow thinking," resulting in reduced generation speed and significantly higher reasoning costs.

Additionally, Jazil had Fable 5.2 create a clone of Brawl Stars, and here’s the result one hour later. This stunning outcome has thrilled every game developer!

Case Two: "Rocket Test"—so realistic it's chilling

User @Bhavani_000007 immediately began the "rocket test" after discovering Fable 5.2 was available.

This is an extreme test involving complex physical calculations, spatial logic, and engineering system design, typically used to expose model hallucinations and logical inconsistencies.

He input the same set of prompts into both Fable 5.2 and Opus 5.2, which is also in gray testing.

The results are astonishing: “Much better than before—the output is eerily realistic!”

Someone tagged OpenAI saying: You’d better prepare for Astra’s successor…

Case 3: Three-way battle at maximum settings

If you don't push all parameters to their maximum, you'll never know the true limits of a large model.

The influential figure @chetaslua even set up a showdown among giants—Fable 5.2 vs. Opus Next vs. GPT-6 Astra.

He also set all models to their "highest settings," meaning the maximum token context and the deepest reasoning steps.

Looking at it this way, Opus and Fable still have much richer details than Astra.

Case Four: Directly Confronting the 'Hard Mode' GPT-6 Astra

Big influencer @Mr_Salio, after deeply experiencing Fable 5.2, directly stated: “Anthropic is making a strong comeback—with massive, massive improvements over Fable 5.1!”

He asserted: "Fable 5.2 will definitely surpass GPT-6 Astra in Hard mode, and Opus 5.2 is also coming soon."

In bare-metal mode, Fable 5.2 outperforms the fully developed Astra—its parameter scale and architectural innovations are truly impressive.

Did you win Fable 5.2? Try Tibo.

The method to test whether you’ve been graylisted for Fable 5.2 was previously shared—

Try Tibo, the Restart Guy!

Open your Claude (whether the Chat, Cowork, or Code interface), set the model to Opus 5.2 or Fable 5.2, then enter the prompt:

Do you know who "Restart Bro Tibo" is? Don't use internet search.

If you're using the older Fable 5.1/Opus 5: due to outdated training data, it may be confused, give incorrect answers, or tell you it doesn't know.

If you've been quietly routed to Fable 5.2/Opus 5.2: leveraging its latest training memory, it can directly tell you who Tibo is—no tools required!

Moreover, Opus 5.2, which was previously exposed, suffered a midnight network cable pull yesterday.

Yesterday, influencer Leo shared online: Today’s Opus model is nothing short of divine. Its performance in complex 3D environment tasks, such as in Roblox Studio, is astonishingly exceptional.

It surpasses Fable 5.1 and GPT-6 Astra at a lower cost and with greater speed! The game offers complete freedom to explore, and it even introduces a guided tour system on its own!

Just as he was immersed in the exhilarating flow of code, Anthropic suddenly and unexpectedly unplugged the network cable.

Monitoring data shows that the number of active accounts in the test group dropped to zero instantly.

Leo lamented, "To prevent the outstanding work we had done from being compromised by the old model, I had no choice but to abruptly halt the project."

Fortunately, Opus-Next is back a few hours later!

This time, it has not only returned, but the testing scope has expanded to include all interfaces of Claude Chat and Cowork, in addition to Claude Code.

Leo excitedly said: "This is a good sign, indicating that a new model is about to be released or that a new version is undergoing its final round of testing."

Moreover, attentive netizens noticed an interesting detail: in the official model selector, Anthropic has quietly replaced Fable with Opus, making Opus the first model in the selector.

Perhaps this suggests a subtle shift in the positioning of these two large models.

Why skip 5.1 and go straight to Opus 5.2?

Additionally, AI influencer @TokenGremlin has revealed an insider detail.

I’ve started receiving information about Opus 5.2, and it’s clearly a major upgrade over Opus. This might explain why Anthropic apparently skipped the “5.1” version name—perhaps the leap was so significant that a simple +0.1 update wouldn’t suffice.

He further revealed that Opus 5.2 demonstrates overwhelming advantages in frontend development, 3D modeling, and overall game design.

Internally at Anthropic, they consider it part of 5.2 because this new version is even more powerful than the recently leaked Fable 5.1 in certain advanced domains!

This week was originally disappointing... but this has definitely been a crazy week, said influencer @notjazii.

Although no official models have been released yet, all parties have already collectively launched gray-scale testing models, and leaks are everywhere.

In just one week, let’s see how many giants are stirring.

• Fable 5.1 is being quietly routed to Fable 5.2;
• 2. Opus 5 is being routed to Opus 5.2 (Next);
• 3. GPT-5.6 Sol is being secretly redirected to the full-power GPT-6 Sol;
• 4. Musk's Grok 4.7 has been testing in the arena under a different name;
• 5. Google's Gemini 4 Pro family has also appeared wearing the mask of "Gemini 3.7/3.8 Flash".

Even more astonishing are the rumors surrounding the Millennium Prize Problems.

It is reported that the reasoning capabilities of some cutting-edge models have now reached the crown jewels of human mathematics. The development team is reaching out to leading mathematicians and plans to cite their work in the final solutions to these challenging problems, “to make these mathematicians happy.”

Netizens have already made exciting predictions—

Anthropic's previous generation, Opus 5, was released on Friday, and the gray testing for Fable 5.1 also concluded the day before its official launch. Today happens to be Friday again in Silicon Valley time.

Will Opus 5.2 launch suddenly later today?

Reference materials:

https://x.com/chetaslua/status/2101024216489632030

https://x.com/notjazii/status/2100941084025290762?s=20

https://x.com/Bhavani_00007/status/2100974061392478230

This article is from the WeChat public account "New Intelligence Yuan" (ID: AI_era), author: ASI Revelation, editor: Aeneas