Based on hacker news https://news.ycombinator.com/item?id=45967211
What People Think About Performance
-
Big Improvement: People are talking about a test called ARC-AGI-2. Gemini 3 got a score of 31.1%. This is much better than GPT-5.1, which got 17.6%. Users think this means the AI is smarter and better at solving new problems.
-
The “Pelican” Test: Users tried a fun test. They asked the AI to draw a computer animation of a “pelican riding a bicycle.” Before, AI could not do this well. Now, Gemini 3 can make good animations.
-
Writing Code: People have different opinions. Some say Gemini 3 solves very hard math problems in minutes, while humans take hours. But others say it still makes mistakes on simple tests. Some programmers prefer the AI named Claude because Gemini’s code is too complicated.
-
Audio and Video: The results are mixed. Some users say it is good at listing who is speaking in a recording. Others say it makes things up (hallucinations) or makes mistakes when watching videos, like analyzing a tennis game.
Price and Access
- Input Cost: The price went up by 60%. It now costs 1.25).
- Output Cost: The price went up by 20%. It now costs 10.00).
- Competition: Even though the price is higher now, users noted it is still cheaper than its main rival, Claude Sonnet 4.5
Privacy and Data
- User Data: One user found a text from Google. It suggests Google uses “user data” to train the AI. Some are worried that Google reads their personal emails (Gmail) or files (Drive). Others say this is normal for legal documents and probably does not include business accounts.
Doubts and Questions
- Test Scores: Some users do not trust the official test scores. They think the AI might just memorize the answers instead of actually solving the problems.
- Number of Users: Google said 650 million people use their app. They think Google counts people who see AI in Google Search or on Android phones automatically.
Comparison to Other Companies
- The Big Picture: Most people agree that Google has caught up to OpenAI and Anthropic. It is very good at thinking tasks now. However, for writing code, many people still like Claude better. People think Google is finally using its research effectively.