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

Google has announced Gemini 4 Argon, positioning the model as its next-generation AI for software engineering, long-horizon knowledge work and cybersecurity. The company says Argon delivered top scores on several internal and public benchmarks, and that Google engineers are already applying the model inside the company. Despite the announcement, general availability has not been opened to external users. Google provided a number of specific claims to illustrate Argon’s impact inside its infrastructure. According to Google, the model used fleet-wide telemetry to identify opportunities that reduced memory usage across its data centers by roughly 300 TiB. The company also reported that Argon agents have been used to migrate substantial C/C++ codebases to Rust, noting migrations that include thousands of lines in the re2 and libgav1 libraries and more than 800,000 lines in the Zircon kernel for Fuchsia OS. To support its performance statements, Google released benchmark comparisons. On the DeepSWE v1.1 software engineering benchmark, Gemini 4 Argon scored 77.9 percent, which Google says surpasses models named GPT-6 Astra, Fable 5.1 and Opus 5.5. The company also highlighted Argon’s leading result on the Vals Index economic analysis test and said the model performs strongly on a range of long-horizon tasks. Although outside access is limited, Google announced API pricing and technical limits for Argon. Introductory rates are set at $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95 percent for a limited time. The model will support an expanded maximum output context of 1 million tokens, up from the 64,000-token limit in earlier Gemini releases, a change Google says will let users tackle larger, single-step tasks.