Google Expands Gemini 4 Argon Access Through Paid API

Wider access and deployment terms

Google made Gemini 4 Argon generally available through its paid Gemini API on October 13, 2026, and began a wider rollout across enterprise and subscription products. The release expands access to a model Google introduced on September 30, when it initially limited use to trusted cybersecurity defenders. More organizations can now evaluate Argon in their own systems, but the wider release does not establish a productivity gain.

Developers can access Argon through the paid API and AI Studio. Google AI Ultra subscribers can use it in Antigravity, while AI Ultra and Pro subscribers can use it in the Gemini app and AI Mode in Search, according to Google’s rollout announcement. The API is unavailable on the free tier or through an AI Pro subscription alone. Google has not specified how quickly every eligible user will receive access or what usage limits will apply.

Google identifies the API model as `gemini-4-argon`. Its model documentation specifies a shared input-output context limit of 1,048,576 tokens. Reaching the maximum output length requires a long-decode workflow rather than an ordinary synchronous request. Introductory pricing is $2 per million input tokens and $10 per million output tokens through November 30; Google lists $4 and $20, respectively, starting December 1. Developers estimating deployment costs need to account for that change.

What performance results establish

Argon ranked first in the October 13 Vals Index at 68.90%, ahead of Claude Sonnet 5.5 at 67.04%. Vals reports that other models lead several component benchmarks. A separate Artificial Analysis index placed Argon High eighth. These results describe performance on different sets of tests; neither ranking establishes how Argon will perform across an organization’s production workloads.

Google has reported engineering speedups and security discoveries involving Argon, but those are examples the company selected. The cited evidence does not establish whether other organizations will achieve similar results. Reliability, latency and total cost on independently chosen workloads remain open questions. Paid API access gives organizations a way to test those factors against their own requirements.

Cybersecurity controls and open questions

Google says the earlier restricted rollout used its Fairwind Program for trusted cyber defenders. Standard cyber guardrails accompany broader access, and Google is accepting further Fairwind applications for access without those standard guardrails. Wider availability gives more defenders a chance to assess Argon, while also making the effectiveness of its misuse controls more consequential.

In its safety account, Google reports that Argon solved all 12 of its simulated end-to-end cyberattack scenarios, compared with five for its previous generation. That company-reported evaluation does not show whether real-world defenders will consistently gain an advantage over attackers. The effects of broader deployment on software development, research and cyber defense—and the effectiveness of safeguards at scale—remain unmeasured.