Google promised the Gemini 3.5 Pro in June, but spent the summer trotting out smaller Flash models. Now, Google is ready to cross the border again with the Gemini 4 Argon. The company claims that this new AI offers industry-leading performance in coding, knowledge work, and cybersecurity, but it is not yet allowed to use it.
While most of us will have to wait to try the Gemini 4, Google says the new model is already being widely used by engineers within the company. Argon reportedly used “fleet-wide telemetry data” to help Google save 300 TiB of memory in its data centers. Meanwhile, Argon agents have been working to migrate C/C++ codebases to Rust at Google, including thousands of lines in the re2 and libgav1 core libraries and more than 800,000 lines in the Fuchsia OS Zircon kernel.
Google also comes armed with a number of benchmarks to back up its claims. On the DeepSWE v1.1 software engineering benchmark, Gemini 4 Argon achieves 77.9 percent, which is higher than GPT-6 Astra, Fable 5.1, and Opus 5.5. Google promises similar power in a variety of long-term tasks, highlighting Argon’s industry-leading score on the Vals Index economic analysis test.
This model is still in limited testing, but Google has announced pricing for the API. For a limited time, Argon will offer fees of $2 per million input tokens and $10 per million output tokens, and cached input tokens will be discounted by 95 percent. The company has also confirmed that Gemini 4 Argon will support a much higher production limit of 1 million tokens. That’s more than the 64,000 tokens of previous Gemini models. Google says this allows users to complete more difficult tasks in one step.
