GPT-5: OPENAI AI turns into a web developer
Openai has just launched GPT-5, its latest large model of language that propels Chatgpt. Among the different news, the company has highlighted improved coding capacities. Described as “the best model for coding”, he is now able to generate web applications from a simple written request.
In the GPT-5 presentation video, Openai presented examples of programming made with a single prompt. For one two, Yann Dubois asked the Chatbot to create a tool to help someone learn French, including monitoring daily progress, with a diversity of activities. In a few minutes, GPT-5 was able to generate the code for a web application, directly functional. And if the first answer is not suitable, just restart it to obtain a fairly different application.
Advertisement
GPT-5 generated this web application from a single prompt. © OPENAI
An intelligent and versatile model
According to Openai, GPT-5 is more efficient than its previous model, O3. On the Swe-Bench Verified test, he obtained a score of 74.9 %, against 69.1 % for O3, while being faster and with 22 % less tokens. It would be particularly gifted to rectify bugs, edit the code and answer questions about complex code bases.
For coding, GPT-5 is more precise than O3, with fewer tokens. © OPENAI
GPT-5 is the most intelligent coding model we have used. Our team found that GPT-5 was remarkably intelligent, easy to pilot and that it even had a personality that we saw in any other model. It detects not only complex and deeply hidden bugs, but it can also execute long background agents and several laps to carry out complex tasks, the kind of problem that previously blocked other models.
Extended context window and availability
In addition, it is now possible to control the reasoning effort, and to switch between faster and direct responses, and more thoughtful responses. GPT-5 has notably been optimized to operate in agent coding tools, such as Cursor, Windsurf, Github Copilot and Codex CLA. Openai has also added a new type of personalized tool. Rather than having to go through the JSON to call on tools, it is now possible to do it in simple text. This eliminates errors due to special characters.
This new model has a context window increased to 256,000 tokens, compared to 200,000 for GPT-4O. This remains less than the million tokens of the GPT-4.1 model, but the latter remains available if necessary.
And it is now available for everyone, including free accounts. Openai has indicated a progressive deployment that has started now.
Advertisement
Want to save even more? Discover Our promo codes Selected for you.
