Just now, Gemini 4 Carbon was released, and Google has returned to the top position in the world.

Just now, Google released its brand-new cutting-edge blockbuster model Gemini 4 Carbon, snatching back the world's No.1 position in one go!

Among the 14 core benchmark tables officially announced by Google, Carbon took 12 first places in one go.

On the latest v5.0 Global Intelligence Index list from authoritative third-party Artificial Analysis, Carbon topped the global ranking with 59 points, completely outperforming Opus 5.5 and GPT-6 Astra!

The Arena community blind test results were also released simultaneously, and it won the global first place with a total text score of 1528 points, and the global second place in the total Agent ranking.

Influencers exclaimed: "Google has achieved the greatest Jedi counterattack of the year!"

Some people said bluntly: Maybe Google has already achieved RSI, so Carbon is so outstanding?

Unfortunately, Gemini 4 Carbon is not yet open to everyone, and is only accessible to "national-level security defenders".

It is reported that Google's internal team has observed that Carbon runs continuously for several days in complex code bases, independently troubleshooting bugs and applying patches.

In addition, in the actual test of the Chrome browser security team, its performance in generating high-quality vulnerability patches is more than 10 times stronger than another cutting-edge large model!

It is gratifying that Carbon will maintain the same introductory period price as Argon:

$2 per 1 million input tokens

$10 per 1 million output tokens

Compared with GPT-6 Astra and Opus 5.5, this price can be called an extremely low bargain! But Google has not yet determined when the introductory period will end.

Full details in 3D actual tests, stable performance after days of continuous operation

Now, the Internet is full of amazing actual test demos.

Some people say that they are too surprised to speak when seeing Google return to the top this time.

=

Some people said bluntly: The 3D cities generated by Carbon are more exquisite and natural than those generated by all other large models.

Some review bloggers said they were shocked by the black hole generated by Carbon.

The following is a collection of reviews.

Review blogger Lumina asked it to write a 3D voxel-style Chinese pagoda scene with complete light and shadow interactions at one time.

The same prompt is put into a four-panel comparison, and the gap is obvious.

The Argon pagoda in the upper left corner is thin and rough, while the Carbon in the upper right corner has been comprehensively improved in the proportion of multi-layered overhanging eaves, the texture of water surface reflection and the level of courtyard landscaping, and its completion degree has directly caught up with Claude Opus 5.5.

In the classic SVG pelican riding a bicycle test, Carbon pushed the details of front-end engineering to a new height.

It not only draws a pelican full of details, but also creates a kinematic cadence slider that supports stepless speed regulation from 20 to 100 RPM.

Click the "CAD Engineering Blueprint Inspection" button in the lower right corner, and the entire picture will instantly switch to a dark blue X-ray mechanical skeleton anatomy diagram, clearly marking the connecting rods of the pelican's leg joints and the gear transmission ratio.

Community developer Raphi-2Code also showed off a dynamic demonstration of a peacock spreading its tail.

Many testers unanimously mentioned that Carbon's thinking time is 2 to 4 times longer than Argon, but it no longer perfunctorily handles tasks.

Peter Gostev, head of Arena capability evaluation, even said bluntly: "Carbon is so much stronger than Argon, I really wish they had skipped Argon and released this version directly."

The return of the king: not only a benchmarking champion, but also an all-round warrior

Previously, Google had long been underestimated in front of OpenAI and Anthropic.

But the birth of Gemini 4 Carbon this time completely shattered all doubts.

Carbon won 15 first places out of 17 tests.

First of all, it shows a crushing advantage in comprehensive intelligence.

In the AA Index v5.0 test, Carbon scored 59 points, surpassing Opus 5.5 and GPT-6 Astra. In the Vals Index v2.2, it also took the first place with a score of 70.4%.

Secondly, it presents an invincible combination of long text processing and logical reasoning.

In the GraphWalks (256k to 1M ultra-long context) test, Carbon achieved an astonishing 96.2%! For comparison, GPT-6 Astra only scored 71.8%, and Opus 5.5 only scored 66.8%.

This means that Carbon makes almost zero mistakes when processing massive amounts of information and finding logical connections across documents.

In addition, it is also a leader in science and engineering fields.

In the RiemannBench test for scientific and mathematical fields, Carbon took a leading position with a score of 80.0%, far exceeding its competitors.

It also won the global first place in both the terminal operation and multi-step programming Terminal-Bench 4.0 with a score of 65.2%, and the scientific research terminal evaluation Terminal-Bench-Science with a score of 64.8%.

In Vending-Bench 2, which simulates long-cycle independent operation, Carbon's score is 1.8 times that of Claude Opus 5.5.

Of course, the competition for specific engineering tasks is still ongoing. For example, on the FrontierSWE 2 and OSWorld-2.1 offline subsets, GPT-6 Astra and Claude Opus 5.5 still retain their respective advantages.

On the Arena Agent list, Carbon directly jumped to 2nd place in the world, leaving Sonnet 5.5 Max and GPT-6 Astra Max far behind.

Among them, in the guidance indicator that measures whether the model can understand human requirements and adjust its direction in time when users change requirements or raise objections midway, Carbon ranks first in the field with +18.6%.

The cost of $1.80 per single task is 72% lower than that of the top-ranked Opus 5.5 Max.

On the Text Arena total text list, Carbon topped the global ranking with 1528 points;

On the WebDev Arena leaderboard, Carbon surged by 89 Elo points and broke into the top four in the world.