Gemini 4 Carbon Global Open: Multi-Agent Swarm Mode Launched, Three Major Evaluations Reach the Top, Lower Costs

Chopsticks officially announced: The new generation flagship Gemini 4 Carbon is now open to the world. Paid API developers, enterprises, and Google AI Ultra users can now get started with AI Studio, Gemini applications, and Antigravity.

Chopsticks clearly stated its positioning on X: it is designed for continuous, long-term complex tasks, performing deep reasoning in large software engineering, cybersecurity, and scientific exploration.

Two weeks ago, during its debut, it landed at the top of the Artificial Analysis chart with a score of 59, surpassing Fable 5.2, Opus 5.5, GPT-6 Astra, and GPT-6.1 Sol on multiple lists.

1. The Most Explosive Capability: /swarm Swarm, the Entire AI Team Works the Night Shift

The most shocking breakthrough this time is the flagship feature launched on the Antigravity platform ------ /swarm multi-Agent swarm mode.

With a single prompt, Carbon can lead the entire AI team to work the night shift.

Combined with DeepMind's newly released AlphaProof Code framework, in just four weeks, the AI swarm generated 6.5 million lines of proofs, validating the compiled code of Signal's core encryption ------ which would originally take human experts years to complete.

The swarm's real-world performance in large construction projects is also impressive:

The restoration project of Google's Platform 37 office building included 7 documents and 206 blueprints. Engineers input /swarm and a prompt before bed, and upon waking, a three-dimensional interactive building was generated;

A single prompt replicated the entire construction process of the Eiffel Tower, and even created an open-world game where a pelican rides a bicycle over the rooftops of Mountain View headquarters;

Restored the 61-key mechanical structure of an ancient piano from the 1780s, placed it into a 3D model of the Koerner Hall in Toronto, allowing for real-time reverberation;

Real-time rendering of an underwater scene in Atlantis;

Given a materials science paper, it effortlessly created a set of fabric mechanics simulation tools, displaying the stress changes of 110 fibers in real-time.

2. Three-Dimensional Restoration: Old Photos and Blueprints Become a 3D Lunar Module

The team handed over old photos and blueprints of the "Apollo Lunar Module" to Carbon, which completely restored the entire lunar module in three-dimensional space ------ whether it’s the complete cabin or the dashboard inside, everything is clearly visible.

Real machine testing here also did not hold back:

Using only the general relativity formula, generated the "Gargantua" black hole from "Interstellar";

Real-time simulation of ink diffusion on rice paper;

A single sentence created a 3D development world.

When compared directly with Fable 5.2, the demo effects were similar, but Carbon saved nearly two-thirds of the costs compared to Fable.

3. Rankings and Prices: Three Major Production Tests Sweep the First Place, Costs Even Lower

Key highlights of the evaluation performance:

GraphWalks test scored an ultra-high score of 96.2%;

CursorBench 4.0, Perplexity WANDR, and Zapier swept all first places in three major real-world production evaluations, surpassing Fable 5.2;

In the multimodal knowledge agent benchmark MKA jointly released by Google and Artificial Analysis, Carbon ranked second globally with 54.3%;

In the independent anti-cheating test KillSwitch-Bench by Nim programming language creator Dominik Picheta, Carbon tied with Fable 5.2 for first place globally;

In a real financial research exam covering 23 cutting-edge models by hedge fund giant Balyasny, it took first place overall;

The cloud security unicorn Wiz used Carbon to scan two major mainstream SaaS service providers, uncovering real zero-day vulnerabilities in less than 2 hours ------ what would originally take several weeks to check for security risks was completed in one afternoon, protecting tens of thousands of downstream enterprises before hackers could act.

The pricing is surprisingly affordable: $2 per million tokens for input, $10 for output, and caching can enjoy a 95% discount. While sweeping the first place in three major production evaluations, the average cost per task is more than half lower than that of the second place.

4. A Free Gift Package for Scientists: 10,000 People with Maximum Access for Six Months

Google also announced a plan: to provide free access for up to six months to 10,000 top scientists worldwide.

Covering fields such as physics, chemistry, biology, materials science, and climate science, the goal is to tackle hypotheses that have been "shelved due to complexity" ------ searching for room-temperature superconductors, cracking complex protein folding to cure cancer, simulating plasma turbulence for controllable nuclear fusion.

The scientific foundation of this tool has already been validated: in the CASP17 protein prediction blind test, it helped scholars increase the retrieval volume of difficult viral protein homologs by 40 times.

Google's vision is straightforward: to make "giving a prompt before bed and waking up to receive results" a daily routine for scientists.