Just now, the brand-new Gemini 4 Pro was fully unveiled, and its real-world test performance completely outperforms Opus 5.5.
Gemini 4 Pro, new checkpoints have been exposed again!
Recently, two brand new checkpoints (internal codename "barium-b" or Checkpoint 2) have been discovered by developers. After actual testing, it is found that Gemini 4 Pro is terrifyingly powerful!
Some developers exclaimed: "After 400 hours of non-stop hard work using Opus 5.5 and GPT-6, Gemini 4 Pro still gives me an incredible shock!"
Google's new generation of killer product demonstrates super strong 3D physics simulation capability, code generation capability and long text reasoning capability.
Some developers exclaimed: "After 400 hours of non-stop hard work using Opus 5.5 and GPT-6, Gemini 4 Pro still gives me an incredible shock!"
Moreover, the attitude of Google's senior management is also very clear.
Koray Kavukcuoglu, the new head of Google DeepMind, publicly confirmed for the first time that Gemini 4 has entered the early stage of post-training.
Its training progress has been greatly advanced, and the early pre-training has been basically completed in only two months.
Get ready, a large wave of hardcore practical tests is coming!
Core Practical Test Highlights: 7 Hardcore Demos
Overnight, it seems that everyone can use Gemini 4 Pro now.
Judging from the shared demos, its performance in 3D generation, physics engine simulation and long text code construction is amazing.
The peacock SVG test is also released, showing a huge leap forward.
Seaplane Takeoff, Crushes Opus 5.5 Completely
Famous reviewer @AI_Screening asked it and Opus 5.5 to generate a physics simulation of a 3D floatplane taxiing on the water and taking off with code.
Previously, the video generated by Opus 5.5 amazed a large group of people, but unexpectedly, this time Gemini 4 Pro's performance completely surpassed it!
It can be seen that although the plane generated by Opus 5.5 is moving, the fuselage shakes violently as if it is about to disintegrate at any time, and the physical feedback of the water surface is also very stiff.
But Gemini 4 Pro runs much more smoothly, with steady acceleration, clear water surface reflection, and very realistic rendering of the "splash" effect.
From the perspective of hydrodynamics logic and picture clarity, it completely beats Opus 5.5.
Generate 3D International Space Station Manually in 16 Minutes
Developer @thtbee_ threw out a blockbuster result.
Without importing any 3D models, Gemini 4 Pro spent only 16 minutes generating an interactive simulation program of the 3D International Space Station orbiting the Earth purely through code!
Surprisingly, it not only built the space station model with Three.js by itself, but also smartly grabbed the correct Earth texture from the Internet to ensure that the latitude, longitude and color are absolutely accurate.
This kind of "creating something out of nothing" modeling ability is amazing.
Of course, some people also complain that the generated UI interface is a bit ugly, and there are still perspective bugs on the back.
From Lighthouse to Rocket Launch (Three.js Rendering)
Netizen @HarshithLucky3 gave it a very difficult instruction: "Use Three.js to turn a lighthouse into a rocket launch, without any UI elements."
As a result, Gemini 4 Pro perfectly followed the instruction!
The tester highly praised its "aesthetic sense": it added textures with excellent texture to all elements in the picture, and the overall 3D output quality is incredibly good.
High-precision Restoration of 3D Wii Remote
Developer @LuminaBench tested its industrial design modeling ability, asking it to generate a 3D Nintendo Wii remote controller.
As a result, it has a very strong grasp of the details of 3D objects, and the generation speed is "unbelievably fast"!
If there is anything imperfect, it seems to over-process the prompt, trying its best to enrich the details even if you don't ask for it.
However, its design aesthetic performance is not very stable on different parts, some parts are excellent, and some parts are slightly out of place.
According to tests, Checkpoint 2 has excellent detail density, much higher accuracy, and it is currently the fastest cutting-edge model that no one can match.
Mechanical Hummingbird, Crushes GPT-6 Sol
Blogger @TimJayas used his long-tested mechanical hummingbird prompt to test the model's understanding of complex mechanical structures and dynamic behaviors.
The conclusion is: "Gemini 4 Pro's performance is absolutely amazing! To be honest, the quality is on par with Fable, but far better than GPT-6 Sol!"
Of course, it is also outperformed by GPT-6 Astra sometimes.
Generate Complete Website in 14 Minutes, Generate Game with Single Prompt
Moreover, its engineering capability is extremely strong.
A developer spent only 14 minutes generating a complete highly creative product display website with full front-end and back-end coverage. "It really only takes a few minutes to