ChatGPT-5: Atomic Bomb or Dud? Practical Applications of OpenAI’s New Model.
In short, the new ChatGPT-5 model represents a huge leap in quality. Although it still operates in a similar way to older models and continues to make relatively simple mistakes, many people believe that the release of ChatGPT 3.5 (the first commercial version) was more revolutionary, and that the introduction of version 4o was also a more significant event.
The new model operates faster, especially when writing texts, which is a considerable advantage, as ChatGPT was known as the slowest model. Despite having a paid version of ChatGPT, I sometimes used the free Grok, which provided fairly good answers but was faster.
Simplifying model selection is also a considerable convenience (currently, the model chooses which tools to use on its own, so we don’t have to wonder whether to use 4o, o3, o4-mini-high, etc.). But it is certainly not on the scale of the Mahnattan project (which Sam Altman spoke about).
Chat GPT 5
Table of contents
-
Vibe coding. Run the program directly in ChatGPT (in Canvas)
-
Summary (Chat GPT 5 Is a Significant Change, but Not a Revolution)
ChatGPT 5 – a quick description.
ChatGPT 5 is the latest generation of the language model created by OpenAI, officially introduced on August 7, 2025. It is a version that combines experience from previous models and introduces a range of improvements, making conversations with it more natural and the results of its work more accurate and useful. No need to choose from multiple models makes it easier to implement AI in a company. The model is available both to free users and to subscribers of paid plans, although usage limits apply to the free version. In addition to the standard edition, mini and nano variants are available, allowing computing power and response time to be adapted to different needs, from simple questions to complex analyses. In the free version, we may not have a clear indication that we are using ChatGPT 5.
How to Use ChatGPT 5 in Practice?
Thinking—get the most out of it.
Above all, it is worth trying GPT-5 Thinking. It will provide us with better answers, although the model will also think for longer. Additionally, if we want to obtain more accurate answers, we can indicate in the prompt that we want it to think about the matter at length, write an answer, reconsider it, and rewrite it, etc. This trick worked with older models, and with this one it is particularly useful when a task seems simple to GPT-5 but it makes mistakes in it.
.
If the model selected complex reasoning, and you want a quick response, you can easily speed up the process.
.
Vibe coding. Run the program directly in ChatGPT (in Canvas)
Many people do not realize how practical it is; vibe coding helps solve everyday problems and is not useful only for software developers.
With code, you can create not only simple applications, but also dashboards or even presentations of sorts. It can therefore be a great help, for example, when conducting research and learning. Additionally, if you commission someone to create a program or website, you can easily prepare preliminary prototypes, making it easy to present what you need.
For example, with a single command I created an aesthetically pleasing, sortable table for finding millionaires who are similar to Elon Musk in their visionary outlook and also have a sense of mission.
.
Code could be generated in language models practically from the very beginning. For a long time, however, it had to be pasted into code interpreters (available online), whereas for many months now it has been possible to get a live preview directly in programs such as Claude, Google Gemini, or ChatGPT. Although, to be honest, before using GPT-5 I didn’t know this was possible, even though I had tested the program a little.
Second example: a timeline of the development of AI models. By clicking on a given event, we gain access to more information. Additionally, we can “stretch” this timeline and filter events. Only two prompts were needed for this.
.
Practical tip: sometimes modifying the program works only moderately well, so it is worth asking for the same thing in a new thread. Sometimes the new version does not appear automatically, so if we do not see the changes, it is a good idea to refresh the thread or close it and reopen it.
There is, however, a noticeable difference in quality; below is a Snake game made in ChatGPT 5.
.
And the version below was done by an older model. In this case, it was 4o, but o4 mini didn’t do much better either.
.
Although the improvement in coding quality is not gigantic (you could achieve some nice results with older models if you experimented with them long enough), ChatGPT 5 has crossed an important threshold. It has become a very good tool for quickly creating attractive charts and visualizations.
Text Analysis
A larger context window means that ChatGPT 5 handles longer texts better. It is able to prepare useful summaries of books. From my experience, ChatGPT is better at “understanding” the subject, while Google AI Studio or Gemini are better at fairly mechanical searches through a book.
Murray Rothbard, The Libertarian Manifesto
If we combine text analysis with simple coding, we can achieve amazing results:
.
Other Advantages of ChatGPT 5
One of the greatest advantages of ChatGPT 5 is its full multimodality. The model can understand and generate responses in the form of text, images, audio, and even video (although video generation is currently unavailable in the application). As a result, it can analyze uploaded photos, describe their content, interpret charts, and discuss recordings. The so-called context window has also been expanded, meaning the amount of data the model can take into account in a single conversation session. It can now process hundreds of thousands of tokens, enabling work on very long documents, projects, and transcripts without losing coherence.
Considerable attention was devoted to improving the quality of reasoning. ChatGPT 5 is better at solving problems that require step-by-step logical thinking. The risk of so-called hallucinations, meaning the provision of false information, has been significantly reduced. According to the creators’ data, the number of such cases has fallen by nearly half compared with GPT 4o. The model also avoids excessive flattery and more often informs users about its limitations, which supports more reliable communication.
In terms of practical applications, ChatGPT 5 expands capabilities in programming, content creation, and data analysis. It can generate clean code in many languages, build websites and applications, and improve existing solutions. In creative writing, it better captures human writing styles, making it suitable for creating stories, articles, or scripts. It also has improved health advice mechanisms, although it still reminds users of the need to consult a professional. Additionally, it can act as an AI agent that integrates with email, calendars, and other tools to perform simple tasks such as making reservations or sending messages, always after obtaining the user’s prior consent.
ChatGPT 5 is widely available. Hundreds of millions of users already use the model every week. The free version provides limited access, while the Plus and Pro plans offer higher limits and the fastest responses. The model can also be personalized—you can adjust the tone, the way content is phrased, or the level of detail. As a result, it works well for learning, work, and everyday conversations alike.
Context window
A context window is the amount of information that a language model can process “at once.” A small context window significantly limits the model’s capabilities, because it “forgets” the content of the conversation thread more quickly, and it also has trouble “reading,” for example, an entire book. There are various tricks that make it possible to work around this problem, but a larger context window is always a significant advantage.
.
The ChatGPT 5 context window is 400,000 input tokens and 128,000 output tokens.
https://platform.openai.com/docs/models/gpt-5-chat-latest
.
Power of the GPT 5 model
In the “last exam of humanity” (Humanity Last Exam), ChatGPT-5 performs very well; however, it is not at the top—that position is held by Grok 4 Heavy, which was released a few weeks before GPT-5.
Oto porównanie GPT-5 i Grok 4
GPT-5 z narzędziami plasuje się między Grok 4 a Grok 4 Heavy w teście Humanity’s Last Exam pic.twitter.com/Y7GDHk2F27
— Beff – e/acc (@BasedBeffJezos) 7 sierpnia 2025 r.
I created the following chart in ChatGPT 5; I asked it to combine the two charts above and organize the data.
.
It should be noted, however, that ChatGPT-5 is significantly cheaper and is available in some form on the free plan, whereas Grok 4 Heavy is only available on the most expensive plan, costing $300 per month.
.
The same applies to the “AI intelligence test”; in this case, ChatGPT is behind the o3 model or Google’s best model:
.
ChatGPT 5 in Coding (Test and Benchmark Results)
Wthe test by qodo ai ChatGPT 5 performs best. My experience with quick, simple projects confirms this as well.
.
Reddit users have had similar experiences:
.
Test results https://www.swebench.com/. ChatGPT-5 just behind Claude 4 Opus, but ahead of Claude Sonnet.
Other Important Things About ChatGPT 5
Dr. Maciej Chrzanowskowski in his LinkedIn post compiled a wealth of useful information about the ChatGPT-5 model:
Hallucinations
- GPT-5 has approximately a 45% lower likelihood of hallucinations than GPT-4o.
- In the GPT-5 Thinking version, the risk of hallucinations is up to 80% lower than in the o3 model.
Generation Efficiency
- GPT-5 generates better answers much faster than the o3 model, especially in scientific and reasoning tasks.
Reliability
- In multimodal tests (after the image was removed), the o3 model gave incorrect answers about nonexistent images in 86.7% of cases, while GPT-5 did so in only 9%.
Generation quality
- Mathematics (AIME 2025, without tools): 94.6%
- Coding (SWE-bench Verified): 74.9%
- Multilingual reasoning (Aider Polyglot): 88%
- Multimodal (MMMU): 84.2%
- Health (HealthBench Hard): 46.2%
An infographic excerpt from the aforementioned post. I encourage you to check out the full post.
Summary (ChatGPT 5 is a significant change, but not a revolution)
GPT-5 was long presented as a gigantic revolution. Just two years ago, it was supposed to be something like AGI, or artificial general intelligence—something capable of operating completely autonomously. In practice, it is simply a better AI model, more convenient to use.
Additionally, due to the number of ChatGPT features, the advantages of a better model are multiplied.