Gemini 4 Argon goes generally available, tops the Vals Index

Google made Gemini 4 Argon generally available on 13 October through the paid Gemini API, AI Studio and Vertex AI, in the Gemini app for Pro and Ultra subscribers, and in Antigravity for AI Ultra subscribers. It was first released on 30 September only to trusted cyber defenders. SVP Koray Kavukcuoglu said it ranks first on the Vals Index for coding, finance and legal work.

Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. Some sources say this runs until 30 November before rising to $4 and $20. Argon scored 68.90% on Vals, ahead of Claude Sonnet 5.5 at 67.04% and Claude Opus 5.5 at 66.97%. Google says agent-submitted code inside the company rose more than 35%.

Tokens are the units models read and write, and pricing follows them. Caching lets a model reuse repeated content cheaply. The Vals Index measures real work tasks in finance, coding, legal and tax, not general knowledge.

Argon can produce up to 1 million tokens in one go via a long decode mode; the default limit is 65,536. A new Gemini app task mode connects to more than 40 apps including Instacart, OpenTable, Adobe and Shopify.

Argon's introductory price is well below rival top models, reflecting the cost advantage of Google's own TPU chips. Running mostly on in-house silicon gives Google more pricing flexibility than rivals that depend on Nvidia, which can add up to real savings for apps processing millions of tokens a month.

Argon arrived after months of delay, with Google skipping Gemini 3.5 Pro. Its aggressive pricing shows Google chasing enterprise share as model gaps shrink to a few points.

Test new models on your own ten real tasks rather than trusting leaderboards, cache long fixed instructions to cut costs, and budget on the permanent price rather than the introductory one.