Newly Intelligent Yuan Report
Google's strongest bomb model so far, fully released!
Pichi officially announced, the new flagshipGemini 4 Carbon,now open to the global audience!
Paid API developers, enterprises, and Google AI Ultra users can now start experiencing it by openingAI Studio, Gemini app, and Antigravity.
Two weeks ago, . At that time, itjumped to the top of Artificial Analysis with a score of 59..
On multiple lists,Gemini 4 Carbon surpassedFable 5.2, Opus 5.5, as well as GPT-6 Astra, GPT-6.1 Sol..
In the GraphWalks test, Carbon achieved an ultra-high score of 96.2%. In the three major real production evaluations of CursorBench 4.0, Perplexity WANDR, and Zapier, it swept all first places in one go, outperforming Fable 5.2, with its initial API pricing as affordable as GPT-6.1 Sol!
The team handed over old photos and blueprints of the "Apollo Lunar Module" to Carbon, which fully reconstructed the entire lunar module in 3D space!
Whether it's the complete module or the internal dashboard, everything is clearly visible, extremely impressive.
The process of 3D reconstruction of the Apollo Lunar Module
More impressively, Google's flagship feature on the Antigravity platform—the/swarm multi-Agent swarm mode.
With a single prompt, Carbon can have the entire AI team work overnight!
The true path to ASI
Google's agent swarm fully embodies the concept of collaboration among small agents.
The most impressive breakthrough this time was its cooperation with DeepMind's newly released AlphaProof Code framework, generating 6.5 million lines of proofs in just four weeks to verifySignal core encryption's compiled code!
And this originally required human experts to spend years to complete.
The real-world measurement of large-scale construction projects is also astonishing.
This time, the swarm of intelligent agents completed the reconstruction project of Google's Platform 37 office building, including 7 documents and 206 drawings.
Engineers input /swarm and a prompt before bed, and it generates a 3D interactive building!
With just a single prompt, the Carbon swarm replicated the entire construction process of the Eiffel Tower.
It reconstructed the 61-key mechanical structure of an 18th-century fortepiano and placed it into a 3D model of the Toronto Koerner Music Hall, allowing real-time reverberation to be heard.
It can real-time render large sections of the underwater world of Atlantis.
One-hand real device extreme testing, directly outputting black holes and my world
We also started real device testing immediately.
It purely relies on the general theory of relativity formulas to generate the "Kagatuya" black hole .
It can also real-time simulate the diffusion of ink on xuan paper.
Then, we increased the difficulty again, letting it create a 3D development world with just one sentence.
Multi-Agent swarm on night shift, single task cost drops 6 times
Currently, Carbon costs only $2 per million input tokens and $10 for output, with cache enjoying 95% discount.
While taking first place in the three major production evaluations, its average single task cost is more than half lower than the second place!
In the multi-modal knowledge Agent benchmark MKA released by Google in collaboration with Artificial Analysis, Carbon ranks second globally with a high score of 54.3%.
Dominik Picheta, the creator of the Nim programming language, just released an independent anti-cheat ranking test KillSwitch-Bench, in which Carbon ties with Fable 5.2 for first place globally.
In the real financial research and investment exam covering 23 cutting-edge models by Balyasny, the top hedge fund giant, it also takes first place.
Cloud security unicorn Wiz used Carbon to scan two major SaaS service providers, uncovering real zero-day vulnerabilities within less than two hours.
Originally, it would take weeks to check security risks, but now everything can be handled in an afternoon. With this, thousands of downstream companies are protected before hackers can act!
"Free Big Package" for global scientists
This time, Google announced an exciting plan: providing up to six months of free access to the top 10,000 scientists worldwide!
Whether they are searching for room-temperature superconducting materials, solving complex protein folding to cure cancer, or simulating plasma turbulence in controlled nuclear fusion, scientists need a "thinking and reasoning research partner."
By offering six months of free highest-level access, Google hopes to enable these 10,000 scientists in fields such as physics, chemistry, biology, materials science, and climatology to tackle hypotheses that have been put aside due to their complexity.
In the CASP17 protein prediction blind test, this tool has already helped scholars increase the retrieval volume of difficult viral protein homologs by 40 times.
This time, Google wants to make "getting a prompt before bed and reviewing the results upon waking" a daily routine for scientists!
References:
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/https://x.com/GoogleAIStudio/status/2120187576489222302
https://x.com/GoogleDeepMind/status/2120219371473617099Light up the star, lock in New Intelligence's ultra-fast delivery!
https://x.com/GoogleAI/status/2120187753329791341
Editor: Editorial Department
Catch up with ASI in seconds
⭐Like, share, and watch all in one click⭐
Light up the star, lock in New Intelligence's ultra-fast delivery!
