Google launches Gemini 4 Argon as its new frontier model
The model outscores Claude Opus 5.5 and GPT-6 Astra on DeepSWE, with 1M token output limit
Google has unveiled Gemini 4 Argon, calling it the next era of frontier intelligence and positioning it for deep reasoning across complex, long-horizon workflows. The launch follows the cancellation of Gemini 3.5 Pro and a shift in focus toward 3.8 Flash, with Argon introducing a new naming scheme alongside substantial capability upgrades.
Argon's output token limit jumps to 1 million tokens, up sharply from 64,000 in prior models, enabling longer and more complex tasks in a single session.
Google says that added headroom lets the model think more deeply and generate hundreds of thousands of tokens in one continuous trajectory, adding a new layer of depth when working through difficult problems.
For the DeepSWE v1.1 coding benchmark, Argon achieved a score of 77.9%, followed by Claude Opus 5.5 with 74.2%, then GPT-6 Astra with 74.1%.
Additionally, on Zapier's AutomationBench, which tests the end-to-end execution of key business functions, Argon performed best with a score of 51.3%. Also, it has state-of-the-art performance of 91.7% on LVBench that tests long video understanding.
Argon was specially designed by Google for cybersecurity defence tasks, with Google claiming substantial improvements over 3.8 Flash Cyber.
The model is going to be released without any cyber guardrails to trusted defenders and Google teams for them to access the model's complete abilities in terms of cybersecurity defence, which will allow Argon to tie for first place on CWE-bench v1, where it scores 68%.
Before wider release, Google says it's focused on defending against misuse tied to cyber and CBRN threats, hardening resistance to indirect prompt injection attacks, and isolating sandboxed environments before high-risk training runs.
Monitoring technology has been implemented by the company to monitor the chain of thought and actions of Argon, which can pause its actions if necessary, without retraining the system with the feedback gathered from the monitoring.
Google Gemini 4 Argon pricing and availability
Google Gemini 4 Argon will be available shortly to Google AI Ultra users and paid API users. The product had earlier been made available only to trusted testers and cyber defenders via the Google Fairwind programme.
The introductory price is pegged at $2 per million input tokens and $10 per million output tokens. The price will go up to $4 per million input and $20 per million output tokens after the introductory period.
-
WhatsApp rolls out cloud backup storage to Android and iPhone users
-
GTA 6 players face PS5, Xbox suspensions for sharing explicit gameplay clips
-
Fruit Battlegrounds codes for November 2026 to enjoy exciting free gems
-
Jensen Huang's potential Nvidia comeback sparks debate amid ongoing controversy
-
China has come forward to voice their concerns after FCC's telecom ban
-
Apple iOS 27.3 update: New features, release date and major changes
-
Dragon Ball Legends Promo codes for November 2026 for improved gaming experience
-
GTA 6 faces mixed reactions as players review story and graphics