Market ·
Gemini 4 has arrived: what Google's new top model changes for agents
Google launched Gemini 4 Argon on October 13, 2026, focusing on long tasks, confirmation before irreversible actions, and resistance to tampering.
By Lucas Marchon, Founder at BrainChat
Google launched Gemini 4 Argon on October 13, 2026, available via API for businesses and in the Gemini app for Pro and Ultra subscribers.
Highlights
- According to Google, the model was trained to ask for confirmation before taking irreversible steps.
- It also became more resistant to prompt injection: in the test cited by Google, the rate of successful attacks dropped to 0.7%.
- A 1 million token limit combining input and output.
- Launch pricing until 11/30/2026: $2 per million input tokens and $10 for output. Afterward, $4 and $20.
For customer service agents
Confirming before acting and resisting malicious messages are critical points for an agent that queries tables, charges, and transfers conversations. Since the cost is much higher than the Flash line, the most common combination is to use the top model for sensitive tasks and Flash for day-to-day volume.
Sources: Google announcement and Gemini API changelog.