Google announces Gemini 4 Argon AI model, but you can't use it yet

Google recently unveiled its new AI model, Gemini 4 Argon, but users won't have access to it just yet. The company showcased the model's potential in various domains, including coding, knowledge work, and cybersecurity, asserting that it delivers industry-leading performance. However, current limitations prevent users from interacting with Gemini 4.

While most individuals will need to wait for the opportunity to test this AI, Google insiders are already utilizing the model extensively. One impressive aspect of Gemini 4 Argon is its ability to leverage fleet-wide telemetry data to help reduce memory consumption within Google's data centers by an estimated 300 TiB.

Moreover, the new AI model is making waves in Google's codebase migration efforts. Gemini 4 Argon is actively aiding the transition from C/C++ to Rust across numerous Google projects, including thousands of lines in the core re2 and libgav1 libraries and over 800,000 lines in the Fuchsia OS Zircon kernel.

To substantiate its claims, Google has provided a series of benchmarks showcasing Gemini 4 Argon's capabilities. On the software engineering DeepSWE v1.1 benchmark, the AI model achieved a score of 77.9 percent, surpassing competitors such as GPT-6 Astra, Fable 5.1, and Opus 5.5. Furthermore, Gemini 4 Argon has demonstrated industry-leading performance in the economic analysis Vals Index test, underscoring its potential across long-horizon tasks.