XNews

Gemini 3 Performance and User Feedback

Gemini 3 Performance and User Feedback

Multiple users report discrepancies in the performance of Gemini 3.0 Pro regarding 'thought partner' capabilities and search quality. - Yuchen Jin reports that Gemini 3 Deep Think surpasses Gemini 3 Pro. - Provided benchmark scores: Humanity’s Last Exam 41%. - Provided benchmark scores: ARC-AGI-2 45.1%. - Some argue that Google has regained its AI leadership since the introduction of Transformers (citing the above scores). - Many users consider Gemini 3.0 Pro an excellent 'thought partner'. - User examples: high-quality thought guidance and responses obtained through deep-think prompts. - Many users list Gemini Pro advantages: multimodal capabilities, ability to directly read YouTube links. - Many users list Gemini Pro advantages: long context support, output length, integration with Google search. - Some users point out that Gemini's search and response still suffer from serious hallucination issues. - Some users believe the quality of Gemini search is lower than ChatGPT Search, with confusion between Deep Research and Deep Think functionalities. - Some users who use Gemini extensively still prefer Codex/ChatGPT subscriptions for coding. - Some users mention subscription choices: Gemini App experience is inferior to AI Studio, and Gemini 3.0 Pro still has a free tier. - Some users provide examples: Gemini can be used to integrate complex central bank and fiscal knowledge points into animated explanations.

Key insights

Gemini 3/Deep Think Rated as Excellent 'Thought Partner' by Users

Many users express that Gemini 3.0 Pro or its Deep Think mode excels in guiding thought, engaging in deep discussions, and generating long texts. User examples include obtaining high-quality, structured thought evolution suggestions using the simple prompt 'deep-think this'; some users, based on years of LLM practice, rate Gemini 3.0 Pro as the best thought partner and recommend daily conversations to enhance thinking efficiency.

Benchmark Data and Performance Claims

Some users cite test scores to support Google's leading assertion: specific benchmark data is referenced by users to assess Gemini's performance: Humanity’s Last Exam score 41%, ARC-AGI-2 score 45.1%, asserting that Google has regained its lead in AI since Transformers. This assertion is based on user-published scores and model comparisons.

Search and Retrieval Functions Still Have Hallucination and Quality Issues

Multiple users noted in comparative tests that while Gemini 3.0 Pro has improved search capabilities compared to previous versions, it still suffers from serious hallucinations, with search result quality deemed lower than ChatGPT Search. Users described specific phenomena where Gemini confuses Deep Thinking mode with Deep Research functionality, leading to incorrect answers.

Specific Advantages of Gemini Pro: Multimodal, Long Context, Google Search Integration

Several user feedbacks highlight the practical advantages of Gemini Pro: multimodal support (including the ability to directly read YouTube links), long context and longer outputs, integration with Google search, good writing performance, and fewer usage limits. Some users claim these features make Gemini App superior to alternative tools in certain scenarios, though others prefer using AI Studio or other models for specific tasks.

Tool Choices and Subscription Decisions: Users Mix Multiple Large Models and Services

User feedback reflects a mixed subscription behavior: some users indicate they cannot fully stop subscribing to ChatGPT and Claude while using Gemini the most daily; others refrain from subscribing to Gemini Pro because they find Gemini App inferior to AI Studio. In coding, many users still prefer using Codex or shared ChatGPT subscriptions.

Gemini's Attempts at Visualizing Complex Knowledge and User Experiences

Some users have attempted to integrate complex financial and monetary policy knowledge points into the same animation or visual narrative using Gemini 3, expressing that the model performs well in such tasks. This feedback reflects users exploring the use of Gemini for high-complexity knowledge integration and presentation.

Source posts