Google has unveiled Gemini 4 Argon, its latest flagship AI model, emphasizing significant advancements in cybersecurity capabilities. The model is initially being made available to vetted security teams through the Fairwind Program, with a phased public rollout to follow.
Gemini 4 Argon has demonstrated strong performance across key benchmarks. On the DeepSWE v1.1 test, which assesses AI performance in software engineering tasks, Gemini 4 Argon achieved a score of 77.9%. This score surpasses previous models, as Gemini 3.6 Flash recorded 49% on the same test in July. For comparison, Claude Opus 5.5 scored 74.2% and GPT-6 Astra scored 74.1% on the DeepSWE v1.1 test.
The model’s cybersecurity features are a primary focus. On Gray Swan’s Indirect Prompt Injection benchmark, designed to test an AI’s susceptibility to malicious instructions embedded within content, Gemini 4 Argon achieved a low score of 0.7%. This indicates improved resilience compared to competitors such as Claude Opus 5.5 and Claude Fable 5.1, which scored 1.0%, and GPT-6 Astra at 8.5%. Google also reports that Gemini 4 Argon attained a 70.9% score on the Wiz Penetration Test Benchmark, outperforming Gemini 3.8 Flash Cyber’s 58.2% score.
Google states that Gemini 4 Argon can process up to 1 million tokens in a single reply, a substantial increase from the 64,000 tokens of previous models. The company also noted that Argon assisted the security firm Wiz in identifying a critical flaw within healthcare software.
A notable aspect of Gemini 4 Argon’s release is its operation “without cyber guardrails.” Google’s stated rationale is that defenders require AI models capable of emulating attacker tactics to identify and patch vulnerabilities before they can be exploited. Acknowledging the dual-use potential of these capabilities, Google indicated that a phased rollout is considered the safest approach. The company is also participating in the U.S. government’s voluntary process for pre-release model access.
The Fairwind Program, Google’s initiative for controlled access to cyber defense tools, launched on September 2 with over 650 partners, including government entities and critical infrastructure operators. This program serves as the initial access channel for Gemini 4 Argon.
Broader access to Gemini 4 Argon will commence with paid API customers and subscribers of Google AI Ultra. Introductory pricing is set at $2 per million input tokens and $10 per million output tokens. Standard rates will be $4 per million input tokens and $20 per million output tokens, with Google yet to specify the end date for the introductory period.
This release occurs within a competitive AI development landscape, marked by recent introductions of new models from competitors like Anthropic and OpenAI. Google’s leadership has also signed a voluntary AI accord previously unveiled by President Trump.
Why This Matters
The materials describe a narrow update: Google has launched Gemini 4 Argon, its most capable AI model, which demonstrates advanced performance in coding, office tasks, and particularly in cybersecurity. The exact end date for introductory pricing of Gemini 4 Argon.
Broader Context
Source materials place the factual news in this context: Gemini 4 is finally here, one week after the release of Claude Opus 5.5 and one day after GPT 6.1 Sol, proving American labs are very much committed to slowing down AI development. Please excuse our sarcasm.
