Google DeepMind has officially unveiled Gemini 4 Argon, its newest frontier model built to tackle highly complex, long-horizon tasks. Optimized for deep reasoning, Argon is designed to handle advanced software engineering, enterprise-level knowledge work, and cybersecurity defense.
The tech giant is utilizing a phased rollout strategy, granting early model access to trusted cyber defenders through its Fairwind Program and working alongside the U.S. government's voluntary pre-release safety framework.
The 1-Million Token Breakthrough
Argon’s most disruptive feature is its massive scale expansion. Google has increased the model's output token limit to an industry-leading 1 million tokens, up from the previous limit of 64,000.
This expanded ceiling allows the model to generate highly detailed, multi-step trajectories. When an AI has the computational headroom to "think" deeply across hundreds of thousands of tokens in a single execution, it can solve massive engineering, legal, or financial problems in one go. Internal Google teams are already utilizing Argon to accelerate codebase migrations, debug daily workflows, and write high-quality technical documentation.
Top-Tier Performance Across Benchmarks
Argon is setting new standards on several industry benchmarks, outperforming previous models in specialized disciplines:
- Software Engineering: It achieved a record-breaking 77.9% score on the DeepSWE v1.1 benchmark, which measures capability in real-world, long-horizon software engineering tasks.
- Business & Workflow Automation: It secured the top spot on Zapier’s AutomationBench with a 51.3% execution score.
- Multi-Domain Economics: The model leads the Vals Index, which evaluates economic impact across finance, coding, legal, and tax work.
- Long Video Analysis: It achieved a state-of-the-art score of 91.7% on LVBench for complex video understanding.
Autonomous Cybersecurity Defense
Beyond coding, Google has specifically trained Gemini 4 Argon to act as a formidable tool for cybersecurity. The model can autonomously locate, validate, and patch critical software vulnerabilities.
To help defense teams utilize its full capabilities, Google is releasing Argon to trusted security partners without traditional cyber guardrails. Cybersecurity firm Wiz has already deployed Argon through its Scan for Good initiative. In early testing, the model successfully uncovered a critical security vulnerability exposing sensitive hospital data that previous frontier models had missed.
Pricing and Safety Rollout
To mitigate risks, Google is implementing multi-layered safeguards against prompt injections, misalignment, and malicious use.
Gemini 4 Argon will launch with an introductory price of $2 per million input tokens and $10 per million output tokens (with a 95% discount for cached input tokens). Following the introductory period, standard pricing will adjust to $4 per million input tokens and $20 per million output tokens. The model will first roll out to paid API customers and Google AI Ultra subscribers before wider developer and enterprise availability.
