Google announces Gemini 4 Argon AI model, currently unavailable for public use

<p>Google announced the Gemini 4 Argon AI model, which follows the earlier Gemini 3.5 Pro promised in June. The company has been testing smaller Flash models over the summer before introducing Argon, which is claimed to offer industry-leading performance in coding, knowledge work, and cybersecurity. However, public access to the model is not yet available.</p><p>Currently, Google engineers are utilizing the Gemini 4 model internally. Reports indicate that Argon has helped save 300 TiB of memory across Google's data centers by using fleet-wide telemetry data. Additionally, Argon agents are involved in migrating C/C++ codebases to Rust, affecting thousands of lines in core libraries such as re2 and libgav1, as well as over 800,000 lines in the Fuchsia OS Zircon kernel.</p><p>Google has provided benchmarks to support its performance claims. On the DeepSWE v1.1 benchmark for software engineering, Gemini 4 Argon achieved a score of 77.9 percent, surpassing competitors like GPT-6 Astra, Fable 5.1, and Opus 5.5. The company also highlights Argon's performance in long-horizon tasks, noting its leading score in the economic analysis Vals Index test.</p>