Gemini 4 Argon

Gemini 4 Argon

Google馃嚭馃嚫google/gemini-4-argon

Gemini 4 Argon is Google's flagship reasoning model for software engineering, math and science problem-solving, and multi-step agent workflows, with output of up to 1M tokens versus 64K for earlier Gemini models. Google reports it beating GPT-6 Astra and Claude Opus 5.5 on 12 of 18 published benchmarks, though those results are company-reported.

Released
Oct 13, 2026
Input
AudioFileImageTextVideo
Output
Text
Support
ReasoningTool callingStructured outputs

Best at

The categories where this model ranks highest.

Life, Physical, & Social Science

#1 路 top 1%

Entertainment, Sports, & Media

#1 路 top 1%

Business, Management, & Finance

#1 路 top 1%

Mathematical

#1 路 top 1%

Performance

Median latency and throughput measured across recent requests.

Throughput86 tok/s
Latency5.61s