Google DeepMind has made Gemini 4 Argon generally available to developers, enterprise customers, and paid Gemini subscribers via API. Argon is a native multimodal sparse mixture-of-experts model optimized for multi-step technical tasks across coding, finance, legal, science, and cybersecurity. Built on TPUs using JAX and ML Pathways, the model accepts text, audio, image, and video inputs while generating text outputs with a knowledge cutoff of March 2026.

The headline feature for developers is Argon’s expanded output capability, which allows the model to write up to 1 million output tokens in a single response. The model operates on a combined input and output window capped at 1 million tokens, utilizing a new continuation protocol for extended decodes. Google introduced introductory API pricing set at $2 per million input tokens and $10 per million output tokens, with cached inputs discounted by 95%. Standard pricing will double after November 30, 2026.

According to evaluation data released by Google, Argon achieved an 87% score on the ERQA embodied reasoning benchmark and posted top results across multiple self-reported evaluation rows against competitors like GPT-6 Astra and Claude Opus 5.5. Google noted that safety safeguards were deployed to mitigate identified biological and cybersecurity risks.

Why it matters

  • A 1 million token output limit opens new possibilities for generating massive codebases and detailed documents in single API calls.

  • Competitive pricing dynamics indicate intensifying margin pressure among frontier model providers.

Source: marktechpost.com