Google promised Gemini 3.5 Pro in June, but spent the summer releasing smaller Flash models. Now, Google is ready with Gemini 4 Argon, claiming industry-leading performance in coding, knowledge work, and cybersecurity, but you aren’t allowed to use it yet. Engineers inside the company are already using the model extensively.
Argon used “fleet-wide telemetry data” to save 300 TiB of memory across data centers. Agents have migrated C/C++ codebases to Rust, including thousands of lines in core re2 and libgav1 libraries and more than 800,000 lines in the Fuchsia OS Zircon kernel.
Google backs its claims with benchmarks. On the software engineering Deep SWE v1.1 benchmark, Gemini 4 Argon hits 77.9 percent, higher than GPT-6 Astra, Fable 5.1, and Opus 5.5. It also scores industry-leading in the economic analysis Vals Index test.
The model is in limited testing, but API pricing is announced: $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95 percent. Gemini 4 Argon supports a 1 million token output limit, up from 64,000 tokens in previous models.
Source: Ars Technica · Summarized by HeadlinesBriefing