Google has released Gemini 4 Argon, a model designed for robotics and complex tasks, according to the DeepMind Blog. The model is available to Google AI Ultra subscribers in Google AI Studio, the Gemini API, and Google Antigravity.
The release comes after months of Google trailing rivals in the frontier AI race. The model is positioned against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across benchmarks for coding, cybersecurity, science, finance, and law, according to The Verge.
Gemini 4 Argon scores 87% on the Embodied Reasoning (ERQA) benchmark, outperforming other frontier models in solving physical, spatial problems in 3D environments. Google said the model is a step change in physical understanding, code generation, and visual-semantic reasoning, and designed it for the entire robotics lifecycle, including robot control, real-to-simulation (Real2Sim) reconstruction, physical understanding, and research and self-improvement.
The model is priced at an introductory rate of $2 per million input tokens and $10 per million output tokens, which will double after November 2026. This pricing is significantly lower than competitors like OpenAI's GPT-6 Astra and Anthropic's Fable 5 models, according to The Verge. The announcement does not include independent test results.
Google said it has already integrated Gemini 4 Argon into internal projects, such as finding and patching cybersecurity vulnerabilities, optimizing YouTube's comments section, and debugging drone physics simulators. The model is also available in the Gemini app, delivering frontier reasoning and improved performance across complex tasks, according to blog.google.
The launch follows a limited release to trusted cyber defenders to test for potential risks, according to The Verge.
Google's release coincides with Meta's announcement of Muse Wonder Full, but Google did not comment on competitive positioning.