EN

Is GPT-5 Really Worse Than GPT-4o?

#AI Tools
2025-09-02 01:029 min read

Content Introduction

In this video, the narrator tests and compares GPT-5 against GPT-4, focusing on their performance across various tasks and prompts. GPT-5 received significant backlash since its release, prompting the narrator to conduct side-by-side tests to assess which model performs better in ten different categories. Key findings reveal that GPT-5 often outperforms GPT-4 in creating visual outputs and reasoning tasks, while also exhibiting a better understanding of user prompts. The narrator notes that while GPT-4 might have faster response times in certain scenarios, GPT-5 excels in tasks requiring deeper reasoning and creativity. The overall assessment indicates that GPT-5 holds advantages in various applications, especially when thinking capabilities are utilized. However, there are still noted deficiencies, such as issues with link generation and references, indicating areas for improvement. The narrator concludes that while there are advancements, the performance of the new model shows mixed results compared to its predecessor.

Key Information

  • Testing was conducted to compare GPT5 and GPT40 due to the backlash against GPT5 and the reintroduction of GPT40.
  • GPT5 faced significant criticism after its release, with many users expressing dissatisfaction.
  • The objective was to run both models across ten different categories of prompts to assess their performance.
  • The testing revealed that GPT5 performed better in creating coherent outlines and generated responses that were more satisfactory than GPT40.
  • However, for speed-related tasks and when not requiring reasoning, GPT40 performed better overall.
  • A dropdown in ChatGPT allows users to choose between different operational modes such as 'auto' or 'fast,' impacting response times and strategies.
  • Various categories tested include document creation, vision capabilities, and ideation, leading to different quality outcomes between the two models.
  • Despite GPT5's better responses in some areas, GPT40 showed superior speed and efficiency in generating responses when less complex tasks were required.
  • Issues such as hallucination and accuracy of links were observed with both models, particularly GPT40, which sometimes produced broken links.
  • Recommendations for improvements include enhancing the decision-making capabilities related to reasoning and providing better citation sources.

Timeline Analysis

Content Keywords

GPT5 vs GPT40 Testing

The speaker discusses their experience testing GPT5 against GPT40, highlighting the backlash received by GPT5 and the desire to compare both models side by side across various categories. The testing involved evaluating their performance on prompts to determine if one is better than the other.

GPT5 Backlash

The speaker mentions how GPT5 faced a significant amount of backlash from users, which became a topic of discussion in their previous videos. They noted that viewers expressed being upset with GPT5's performance.

AI Thinking Models

The video outlines the differences in thinking models between GPT5 and GPT40, explaining how these affect their performance and decision-making capabilities.

Model Options

A distinction is made between the various model options available for users, including 'Auto' and 'Fast', with 'Auto' being preferred for a fair comparison.

Customization in AI

Discussion around the customization options available to users in GPT's settings, particularly focusing on personality adjustments for the AI's interactions.

Image Generation

The speaker tests each model's ability to handle image generation tasks and their effectiveness in creating visual content, alongside the discussion on how hallucinations and misinterpretations can occur.

Performance Testing

Through structured prompts, the speaker demonstrates the abilities of both models in generating code, visual content, and solving logic-based tasks, providing a comparative analysis of outputs.

Citation and Trustworthiness

There is a focus on how both models handle citations and generating reliable sources, with GPT40 reportedly producing more accurate links than GPT5 during testing.

Visual Dashboards

The speaker further examines the capability of both models to create visual dashboards from data inputs, noting the differences in presentation and functionality.

General Observations

Overall observations from the speaker reveal that while both models excel in specific tasks, GPT5 appears to perform better in logic and reasoning tasks than GPT40, which occasionally resorts to irrelevant outputs.

Future Expectations

Concluding remarks emphasize the anticipation of continued developments in AI models, particularly in improving reasoning abilities and user satisfaction.

Related questions&answers

What has been tested in the last couple of days?

GPT5 versus GPT40.

What kind of feedback did GPT5 receive?

GPT5 received a ton of backlash with many people expressing their dislike for it.

What categories were used for testing GPT5 and GPT40?

The tests were run across 10 different categories of prompts.

What plans are available for users when using these models?

Users can choose from auto, fast, and pro plans, where the pro option is intended for research.

How did GPT5 perform compared to GPT40 regarding image handling?

GPT5 performed better in some aspects, while GPT40 was found faster overall.

What issues arose during the hallucination testing?

GPT5 was less likely to generate fake URLs compared to GPT40, which had problems creating valid citations.

What was concluded from the coding tests?

GPT5 showed superior capability in creating visual elements and dashboards, while both models faced issues with links and image generation.

How does reasoning affect the performance of the models?

GPT5 excels when reasoning is needed, while GPT40 lacks this capability.

What was a significant takeaway regarding user experience?

The ability to not choose between a ton of different models contributed to a smoother user experience.

What additional tests are planned for these models?

Future tests will include more complex prompts and comparisons with other models like Gemini and Claude.

What overall impression was left about the new models?

While both models have their strengths, GPT5 is preferred for creative and reasoning tasks, while GPT40 shows faster execution for direct tasks.

More video recommendations

Share to: