BREAKING
Gemini 3.5 Flash Gets Computer Use
OSWorld-Verified Scores
GPT-5.5
78.7
Gemini 3.5 Flash
78.4
Claude Opus 4.7
78
Gemini 3 Flash
65.1
0
%
Gemini 3.5 Flash
0
%
Gemini 3 Flash
0
M
Input tokens
How Computer Use Works
1
See the screen
↓
2
Plan the task
↓
3
Click and type
↓
4
Confirm risky steps
Promise vs Concerns
Strengths
●
Fast and cost-efficient
●
Built into mainline model
●
Hardened vs prompt injection
Concerns
●
Tends to overthink
●
Ignores instructions
●
Pricey on hard tasks
Agentic AI Goes Mainline
AI NEWS BLITZ
Google DeepMind has built screen-operating Computer Use right into Gemini 3.5 Flash.