Gemini 4 Argon is Google's first Gemini 4 model and its push back to the frontier, leading several agentic and long-video evaluations (DeepSWE v1.1 77.9%, AutomationBench 51.3%, LVBench 91.7%) and Google's prompt-injection robustness tests. It handles text, documents, charts and video, and raises output length to 1M tokens. Access is restricted for now: it is rolling out to vetted cyber-defense teams through the Fairwind Program while Google tunes guardrails, with general API availability promised at an introductory $2/$10 per million tokens, rising to $4/$20 afterwards.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| SciCode | 61.8 | accuracy | today |
| TerminalBench | 57.1 | accuracy | today |
| LCR | 79.7 | accuracy | today |
| Humanity's Last Exam | 57.1 | accuracy | today |