Google launches Gemini 4 Argon, its new frontier model

Top spot on the Vals Index comes with a per-token price tag to match

Google has released Gemini 4 Argon, a frontier model built for "deep work" on complex tasks. The company names coding, finance, legal, cybersecurity and science as target domains.

It is available now through the API in AI Studio and the Gemini platform. It also reaches Google Antigravity for AI Ultra subscribers, and the Gemini app and AI Mode in Search for AI Ultra and Pro subscribers. Google says Argon ranks first on the Vals Index, a benchmark covering finance, coding, legal and tax work. The company is pricing the model at $2 per million input tokens and $10 per million output tokens at launch, with cached input tokens discounted 95%.

Google is leaning heavily on internal case studies to make its case. The Chrome team reportedly used Argon to separate roughly 6 million lines of legacy browser-feature code from Chromium's 40 million-line codebase. It finished an architectural refactor across hundreds of files in weeks rather than months. On Google's open-source Agent Development Kit, Argon working overnight cut in-memory session prep time by 40% and sped up cold imports by 21%. Those changes are now live in the public repository. Google also says Argon delivered a 2.85x end-to-end inference speedup for its Gemma 4 E2B model on an A100 GPU via vLLM. Throughput rose from about 37,900 to over 108,000 tokens per second, with no drop in output quality. Google says the tuning job would normally take months.

Outside Google, the company points to early-access partners as evidence of real-world impact. Wiz used Argon under its "Scan for Good" initiative to help a federal agency close a data-exposure gap. It also found a public file download bypass that exposed a master key and administrator access to healthcare infrastructure, protecting what Google says were hundreds of thousands of patient records. Isomorphic Labs says Argon compresses weeks of patent and PDF synthesis for drug discovery into hours. Google is also expanding its Fairwind cyber-defence program to hundreds of additional partners. The company cites Argon's use in quantum error-correction work on AlphaQubit, a stellarator-design result extending a 1967 mathematical conjecture, and genome analysis at Gamow Labs that reportedly caught a chromosomal rearrangement other software missed for years.

All of these results come from Google itself or from partners in its early-access program. That includes the Chrome refactor, the Gemma speedup, the Wiz vulnerability finds, and the fusion and quantum work. No independent benchmarks are cited. The claims are specific and operationally detailed, more than most model launches offer. But they remain self-reported wins from parties with a stake in the model's success.

Written from the primary source above by The Neural Net’s automated desk (claude-sonnet-5, theneuralnet/pipeline@1) without human review. How this works. Corrections: check the source first — if we got it wrong, the source link is the record.