The new version of ChatGPT, ChatGPT 5, was released on August 7, 2025.
Presented as a major update, this new version of AI promises more power, fluidity and real-world applications, notably as anAI agent. However, ChatGPT 5 has been the subject of much criticism!
In this article, we compare ChatGPT 5 with its predecessor, GPT-4, and explain when it’s best to revert to the old version, ChatGPT 4o.
We also point out the limitations and criticisms of GPT-5 as reported by users and the press. Finally, we explain the differences and uses between free ChatGPT and its paid models, which also give access to the old ChatGPT 4o version.
You’ll also find all the features of the new version of ChatGPT. Here we list all the differences between its sub-models: GPT-5 Auto, Instant, Thinking and Pro.
What’s new in ChatGPT 5 (vs GPT-4o)?
Open AI’s new model, ChatGPT 5, now combines advanced reasoning, multimodal creation and management using text, images and voice, and the execution of complex tasks, in a single system. GPT-5 becomes the default in ChatGPT and replaces GPT-4o, o3, o4-mini… The former GPT 4o model remains available to subscribers.
No need to manually choose the most suitable model, GPT-5 directly (and dynamically) selects the best model for the task, and therefore the best thought process for your application.
GPT-5 is designed to handle complex logic and provide in-depth reasoning.
Like its predecessors, it’s as good at code as ever! It’s just as good at generating text and images. It can be integrated with external tools.
In fact, GPT 5’s contextual memory is off the charts (over 400,000 tokens), so it has a very good memory for long interactions, and does so consistently. Fewer bugs than previous versions, where ChatGPT tended to get confused and respond to old conversation prompts in its replies to long conversations!
As a reminder, the definition of a token in AI: a token is a text unit used by GPT models to process information. It can correspond to a whole word, part of a word or even a punctuation mark. The more tokens a query contains, the longer it takes to process, and the more resources it consumes (and therefore credits if you’re paying for them!).
What does the new version of ChatGPT offer?
The new generation of ChatGPT introduces major advances. OpenAI has focused on the recurring problems of older GPT versions to correct them. The AI company has focused on the fluidity of exchanges, the power of analysis, but also personal use cases.
In concrete terms, the new features are :
- Intelligent and adaptive reasoning : GPT-5 decides when to respond quickly or think in several steps for complex problems, making dialogue and problem-solving more effective;
- (Very) extensive context management: the AI can store up to 256,000 tokens (units of text) in a ChatGPT session, and even 400,000 via the API! ChatGPT can process and analyze long documents, analyze entire meetings or large databases without losing the thread;
- Revised multilingualism and voice fluency: the new version understands and translates a large number of languages more accurately, and adapts its voice function to accents and intonations, resulting in more natural voice conversations;
- Business application connectors: GPT-5 connects directly with external tools such as CRM, ERP or databases. GPT agents (more details below) can handle complex business tasks without slowing down or adding costs.
What is a GPT agent?
A GPT agent is a specialized, customized version of ChatGPT. It will enable you to perform a specific task in which it is specialized and/or interact with a particular environment or area of expertise.
Compared with the classic version (where the user sends a prompt and receives a response), a GPT agent will be able to follow pre-programmed instructions and is configured to work within a defined framework: customer service, creation of marketing content, data management, and so on.
The GPT agent can connect to external tools: CRM, database, API, Google Sheets, Notion, etc. In concrete terms, you ask a question and the AI queries the database, then gives an answer in the form of a report! It can then create a report page on Notion, for example.
Like its predecessors,ChatGPT can be usedto automate processes. It keeps a local “memory” and will adjust its behavior according to user, domain and objective.
Thus transformed and highly personalized, ChatGPT becomes a fully operational AI assistant, capable of handling concrete, specialized tasks. Even if human supervision and verification are still necessary, not to say mandatory!
How do I configure and use a GPT agent with ChatGPT 5?
One of the most popular new features of ChatGPT 5 is the simplified creation of an AI agent, a virtual assistant, a virtual assistant. To set up such a personalized AI assistant, it is essential to use a chatbot creation platform.
No-code platforms such as Botnation, Manychat or OpenAI Studio make it possible to centralize the configuration of the GPT agent, with its rules, prompts, instructions and limits, then integrate it easily, with efficient monitoring.
Without this, it remains a “crude” AI, unable to interact effectively with your users or your business tools.
Here’s how to set up your own GPT agent, following these steps:
- Define the agent’s mission: product FAQs, making appointments, etc. ;
- Give him precise instructions on his role;
- Provide reference content (documents, lists, links) ;
- Select the right GPT-5 template in the AI chatbot creation platform;
- Deploy the agent on the desired channels (CRM, WhatsApp, Slack, website…).
ChatGPT 5 vs 4: malfunctions?
Many press articles are critical of ChatGPT 5 compared to previous versions, especially GPT 4o!
Criticism of GPT-5 is rife in the AI community, echoed in the press, on several points: conversational capabilities, creativity, personalization and the disappearance of old models.
OpenAI has recognized some of the issues raised by users and has made adjustments, which are still in progress. Notably, OpenAI allowed the return of the old GPT 4o model, in response to media pressure!
The technical malfunctions noted are :
- Loss of creativity and standardized responses: GPT-5 is considered more neutral and repetitive than GPT-4o. The content produced often lacks originality and personalization, limiting its use for rich dialogues or creative work;
- Logical and contextual errors: during complex exchanges or long sessions, the model can lose the thread of the conversation, mix up subjects or produce incoherent reasoning;
- Overly short or sanitized answers: GPT-5 sometimes adopts an overly cautious tone, with formatted, synthesized answers that are ultimately unhelpful.
- Limited personalization: AI would be less emotionally involved, with difficulty in adapting its responses;
- Persistent hallucinations: despite optimizations, factual errors persist, especially when using ChatGPT for automatic routing between its sub-models.
ChatGPT 5 vs 4: what are the differences?
If you’re feeling nostalgic, it’s possible to go back to ChatGPT 4! OpenAI has left version 4o for its ChatGPT subscribers, but it’s not possible to go back to other versions (1o, 3o, etc.).
Many users are debating ChatGPT 5 vs. 4, widely discussed in the AI community about the strengths and weaknesses of each version.
Why do some users prefer GPT-4o to ChatGPT 5?
The abrupt change in ChatGPT’s tone and interaction caused a real sense of loss, if not unease, among users attached to the AI’s “personality”! Notably, in terms of response stability and automation of existing workflows, GPT 4o would be much more popular than the new GPT 5 version.
- Answers are considered shorter and less relevant: many note that GPT-5 provides briefer answers, overall less detailed, especially on complex or creative topics;
- The tone is colder and more impersonal: GPT-5 is criticized for its more formal and “corporatist” style, very different from the more natural exchanges with GPT-4o ;
- So there’s less creativity and contextual engagement: the new ChatGPT model is optimized for productivity, which is detrimental to “conversational” uses, such as coaching,personalized assistance or even… emotional support;
- Despite OpenAI’s promises to reduce errors, some users report an increase in inconsistencies or a lack of logic in certain responses, sometimes exacerbated by automated routing between sub-models;
- ChatGPT Plus subscribers regret the loss of functions (query volume, template diversity), and some workflows optimized on GPT-4o are no longer compatible.
When should I switch back to GPT-4 instead of GPT-5?
We advise you to return to GPT-4o rather than use GPT-5 for AI style, creativity, natural dialogue and personalization. On the contrary, we advise you to use GPT-5 for complex, technical, analytical tasks (such as math, data, logic, scientific synthesis).
Here is our comparison table between GPT-4o vs GPT-5:
| Criteria | GPT-4o | GPT-5 |
| Creativity & style | More expressive, narrative, fluid → ideal storytelling | More neutral and formatted answers, less original |
| Conversational richness | Natural dialogue, warm tone, humor and nuance | Less emotional engagement, sometimes cold |
| Personalization | Good adaptation to tone of voice and context | More standardized, homogeneous tone |
| Logic & reasoning | Correct reasoning but limited to complex steps | Multi-step, analytical and structured reasoning |
| Long answers | Detailed explanations | More concise, sometimes too cautious |
| Multimodality | Text + voice + image (limited) | Text, image, code, external tools, +400k tokens |
| Workflow stability | Well-integrated, stable with existing automation systems | Migration sometimes complex, compatibility to be adapted |
| Ideal use case | Storytelling, creative marketing, human dialogue | Mathematics, coding, analysis, scientific synthesis |
Beyond ChatGPT, don’t hesitate to look at other AIs, even if ChatGPT remains one of the best AIs on the market. See our article comparing AI alternatives to ChatGPT!
What are the differences between ChatGPT 5 models?
ChatGPT 5 offers 4 models, each with its own specific features. Here’s a breakdown of each one and the uses you can make of it: Auto, Instant, Thinking and Pro.
These ChatGPT models are available with a subscription. To use ChatGPT version 4o and not GPT 5, you will also need a subscription. The GPT-5 Pro model is not available with a simple subscription, but requires a pro subscription.
GPT-5 Auto: the complete, multi-purpose model
GPT Auto evaluates the complexity of your prompt in real time and automatically selects the sub-model best suited to respond to the prompt. It then switches to the ideal engine to guarantee the best possible response.
ChatGPT Instant: for instant answers
ChatGPT Instant is all about speed and fluidity. It activates a sub-model specially optimized for everyday use: fast text generation, simple writing, light chat. Here, advanced logic takes a back seat to ultra-low latency and near-instantaneous response. It’s ideal for dynamic conversations, short tasks and all scenarios where reaction time takes precedence over… depth of analysis!
ChatGPT Thinking: for in-depth reasoning
ChatGPT’s Thinking mode enables complex answers requiring structured reasoning or multi-step analysis.
When you need to “take time to think”, this engine increases computational effort and produces more precise, detailed and logically organized answers. This is the mode of reference for solving mathematical problems, building strategic plans, analyzing data or writing reliable code. It will respond in a jerky fashion: in dashes, point by point, analytically and logically.
GPT-5 Pro (intensive parallel reasoning)
This model is available for Pro and Enterprise subscribers. GPT-5 Pro is a calculation mode that enables ultra-precise reasoning, very useful for fields requiring total reliability (such as scientific, legal or medical/financial). It is the most resource-intensive mode, both in terms of computing power and precision.
Free ChatGPT 5 and its paid versions
ChatGPT is still available free of charge to all Internet users! OpenAI offers a number of pay-as-you-go packages, more suited to certain needs.
Free ChatGPT
For standard use, the free version of ChatGPT offers daily limits on the number of messages and the use of advanced templates. We use ChatGPT free for occasional use.
ChatGPT Plus: is a subscription worth the cost?
ChatGPT’s paid subscription is well worth it if you use AI on a regular, even daily, basis. Admittedly, it’s an investment, but a highly profitable one (around 20 euros per month). You get access to improved ChatGPT performance, includingaccess to the older GPT-4 version. We highly recommend ChatGPT Plus for professionals!
ChatGPT Pro for businesses and/or developers
ChatGPT’s Pro subscription offers the maximum power of GPT-5. There’s priority access to advanced modes, more depth in AI reasoning with truly effective speed. ChatGPT Pro has been designed for intensive and complex use. So fewer AI malfunctions!
ChatGPT API rates
If you wish to integrate ChatGPT-5 into your applications, the price depends on the model chosen (Fast, Thinking, Pro) and the volume of tokens exchanged. ChatGPT’s APIs are charged on a pay-as-you-go basis, allowing for total flexibility of use. ChatGPT APIs are essential for developers and companies integrating ChatGPT into their applications and AI agents.
