ChatGPT 5 Has Arrived: What's New and How Does It Stack Up?

The latest iteration of OpenAI's flagship AI model, ChatGPT 5, is now becoming the default experience for all users, including those on the free tier. So, what can you expect from this new generation? Let's dive into the key improvements and areas where it still needs work.

Faster Performance for Everyday Tasks

One of the first things you might notice is an improvement in speed. For lighter queries, such as asking how many rings Saturn has or how many distinct iPhones Apple has made, GPT-5 is significantly faster, performing 30% to 50% quicker than the previous GPT-4o model. While speed wasn't previously considered a major issue, this enhancement certainly contributes to a smoother user experience.

Simplified User Experience: One Model to Rule Them All

Perhaps the most significant change in GPT-5 is its ability to unify the capabilities of previously separate, specialized models into one general model. In earlier versions, users, especially those paying a monthly subscription, faced a confusing array of options like GPT-4o, 03, 03 Pro, 04 Mini, and 04 Mini High, each optimized for different tasks. Now, GPT-5 intelligently selects the best underlying model for your specific question, whether it requires simple information retrieval or more complex "thinking". This means you no longer need to manually choose a specific mode for coding, creative writing, or light questions – the AI handles it automatically.

Enhanced Capabilities for Complex Tasks, Especially Coding

GPT-5 shows notable advancements in handling complex requests, particularly in coding. When tasked with generating a playable Tetris game, GPT-5 produced a "vastly better game" than previous models. While the older GPT-4o took 57 seconds for a basic game and the specialized 04 Mini High took 14 seconds for a similar version, GPT-5 completed its version in 19 seconds. Crucially, GPT-5 demonstrated an "intuition" to add more complex features unprompted, such as score tracking, level progression, showing upcoming pieces, and on-screen controls, resulting in a more sophisticated game with more lines of code.

Designed by Freepik

This "implicit understanding of what makes the game look more premium" was even more evident when asked to create a playable chess game with Pokémon pieces. While previous models struggled or produced non-functional results, GPT-5 took significantly longer, actively stating it was "thinking longer for a better answer". The outcome was a fully functional chess game with features like highlighting legally available moves and indicating whose turn it is, showing it is "leagues ahead of last gen" in this area.

Accuracy and Hallucination: Still a Work in Progress

A long-standing issue with AI chatbots is "hallucination"—their tendency to confidently state incorrect information. While OpenAI claims GPT-5 should be better thanks to more powerful hardware, user feedback integration, and improved benchmarking, the model still "invents stuff" with complete confidence. For instance, when asked for 10 tech products made by food brands that can actually be bought, GPT-5 still generated fictional items like a "McDonald's XT mobile phone" or "KFC gaming console," performing no better than its predecessor in this specific test.

However, GPT-5 does appear to offer a "much more human-oriented answer" when asked about its own identity, successfully discerning the user's intention even when the question was phrased obtusely.

Creative Endeavors: Mixed Results with Writing Flair Improvements

In creative tasks, GPT-5 presents mixed results:

• Image Generation: The sources suggest that GPT-5 is "not a step up when it comes to image generation" and can even be "really bland" or worse than GPT-4o. Examples include generating YouTube thumbnails that were less aesthetically pleasing, incorrectly sized, or contained "cursed" self-images, and birthday party invitations that lacked effort compared to GPT-4o's output.
• Writing Flare: Conversely, GPT-5 shows improvements in "writing flare". When tasked with writing a 30-second "Tech Fail" script about the Windows Phone, GPT-5 produced an answer with analogies that "make you think" and included additional valuable elements like notes for filming and B-roll shots, a feature absent in the GPT-4o response. While "still needs work," it provides a much better starting point and feels less "wooden" than older models which often focused solely on providing an answer rather than its delivery.

Accessibility and Pricing

Despite its advancements, the monthly subscription price for ChatGPT remains the same. For non-paying users, GPT-5 is still accessible, though it comes with a cap of approximately 8 to 10 requests per day using the most powerful version. Once this limit is reached, users will then be shifted to a "less powerful but still more powerful than last generation GPT-5 Mini".

Conclusion

In conclusion, while ChatGPT 5 is "not better at every single thing," particularly image generation and completely eliminating hallucination, it represents a significant step forward in many areas. Its ability to simplify the user experience by integrating complex capabilities into a single model, coupled with improved performance in complex coding tasks and a more natural writing style, makes it "pretty hard to complain about".