Google DeepMind has introduced Gemini 4 Argon, a new large language model designed for coding, enterprise knowledge work, and cyber defense. The model is currently available in a limited cybersecurity preview through the Fairwind Program for government users and trusted defenders.

  • Gemini 4 Argon features an industry-leading 1M-token output limit, achieved via a new "Long Decode Continuation" API that pauses and resumes long responses.
  • Google claims the model achieves state-of-the-art results in 13 of 19 credible benchmarks, outperforming GPT-6 Astra and Claude Opus 5.5 on metrics like DeepSWE.
  • Standard pricing is set at $4/$20 per 1M input/output tokens, with a 50% introductory discount available.
  • Independent evaluations from Artificial Analysis and Vals show mixed results, with Argon ranking highly on coding and security tasks but showing lower accuracy and higher hallucination rates compared to competitors like Astra.

The model is positioned as Google's response to recent frontier models from OpenAI and Anthropic, aiming to re-establish leadership in the AI space.