Gemini 4 Pro: Leaked benchmarks indicate top performance
Google is preparing to launch the new AI model Gemini 4 Pro. After months of radio silence and numerous leaks, everything points to a release soon. The model is intended to outperform the competition in terms of performance and undercut it in terms of price.
Google is pushing AI development
Google is working hard to complete the Gemini 4 Pro language model. After a long period without major AI releases, the company plans to launch the new version before the end of 2026. The model is currently in the post-training phase. Security guidelines are primarily implemented and fine-tuned there. Numerous unconfirmed leaks and reports about alleged test runs of the software have been circulating for a long time.
Benchmark results appeared on various platforms and were attributed to an internal Google project. Such preliminary information suggests performance improvements, but should be viewed with great caution. Isolated tests often do not paint a complete picture of everyday performance under real-world conditions.
Focus on code and 3D design
Like DeepMind boss Koray Kavukcuoglu in a conversation with The Information confirmed, the current focus is on fast and secure deployment. Google had postponed the development of the planned interim version Gemini 3.5 Pro in favor of smaller models. Now the direct jump to the fourth version is intended to reconnect with current competing products on the market.
According to the leaked data, the new model shows particular strengths in software development and three-dimensional content creation. It is speculated that a new architecture will ensure significantly more precise data processing. Developers could benefit from improved error correction and more efficient code generation in the future. In addition, complex graphics can be rendered in a shorter time.
Price war on the AI market
In addition to pure computing power, pricing plays an important role for end users and companies. Unconfirmed documents suggest that Google is pursuing an aggressive price war to regain lost market share. The strategy aims to dramatically lower the barriers to entry for using advanced models. A big advantage of the new generation could be the ability for recursive self-improvement.
This means that the system optimizes its own algorithms completely independently during ongoing operation. One disadvantage that remains for the time being is the lack of transparency regarding the exact data basis. The actual hardware requirements are also still unknown and will probably only be clarified at the official launch.
