Google Unveils Gemini 4 Argon Frontier Model with Enhanced Capabilities
Google has announced Gemini 4 Argon as its latest frontier model, designed for 'deep reasoning across complex, long-horizon workflows'. The new model is built upon a new naming scheme and offers capabilities in areas such as coding, knowledge work, cybersecurity defense, and creative writing. It features an increased output token limit of 1M tokens, allowing for longer and more complex use cases.
Gemini 4 Argon has demonstrated impressive performance in various benchmarks, including a DeepSWE v1.1 score of 77.9%, surpassing other prominent models such as Claude Opus 5.5 and GPT-6 Astra. The model is also highly capable in cybersecurity defense, with Google touting leaps over its previous model, Gemini 3.8 Flash Cyber.
The company has emphasized the importance of defending against misuse from cyber or chemical, biological, radiological, and nuclear (CBRN) attacks, as well as prompt injection attacks. To mitigate these risks, Google is implementing various measures such as monitoring internal activations for misuse, hardening systems through sandboxed environments, and deploying misalignment mitigations.