Google's Argon model cuts Chrome refactor from months to weeks

Google has released Gemini 4 Argon to developers and enterprises, and its own engineers are the first customers it is showing off. The headline example: the Chrome team used Argon to untangle roughly 6 million lines of legacy browser code inside a 40 million line codebase, finishing an architectural refactor across hundreds of files in weeks instead of months. Google says agentic code submissions inside Antigravity rose more than 35% while rollbacks fell by nearly a third. Those are Google's numbers about Google, not independent audits, so treat them as a strong signal rather than a settled benchmark.

The pricing is the part that should get a finance team's attention. The introductory rate is $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off. For companies running large codebases through agents, cheaper cached context can matter as much as raw capability. Argon also claims the top spot on the Vals Index, which covers finance, coding, legal and tax work, though the benchmark is one more vendor-selected yardstick until a non-Google customer reports results on its own workload.

The more concrete examples are smaller and more useful. The Agent Development Kit team reports a 40% cut in in-memory session prep time in long sessions after an overnight autonomous run, and Google says Argon helped nearly triple the inference speed of its open Gemma models on vLLM. Isomorphic Labs says it is using Argon to cut patent and PDF synthesis in drug discovery from weeks to hours, and Wiz is using early access for cyber defense work.

If you run engineering or research at a company without a Google-sized codebase, the lesson is to pick one measurable workflow, such as a refactor or a test backlog, and run Argon against it this quarter. Companies that wait for a second round of case studies will find their competitors have already redesigned the work around it.