Google's Gemini 4 Argon Surpasses Rivals in Cybersecurity and Software Engineering

What happened: Google unveiled Gemini 4 Argon, its latest AI model, on September 30. Argon leads 12 of 18 benchmarks in Google's own comparison, including a 77.9% score on DeepSWE v1.1 (software engineering) and 85.8% on source-code vulnerability discovery. Its Gray Swan Indirect Prompt Injection attack success rate is just 0.7%, outperforming Claude Opus 5.5 (1.0%) and GPT-6 Astra (8.5%). Argon can output up to 1 million tokens per response, a major leap from the previous 64,000-token limit. Initial access is restricted to vetted security teams via the "Fairwind Program," with broader API and subscription rollout to follow. Introductory pricing is $2 per million input tokens and $10 per million output tokens.

Why it matters: Gemini 4 Argon's benchmark dominance signals Google's renewed leadership in AI model security and long-context reasoning, especially for cybersecurity applications. The decision to release the model without cyber guardrails to select defenders is controversial, raising dual-use concerns. However, the move is positioned as a defense-first strategy, aiming to empower cyber professionals with advanced tooling. The limited release and high pricing reflect both the model's capabilities and Google's intent to control access during early deployment.

Source: Decrypt