At Portia AI we have been testing GPT-5 since its release and we announce that it will be available to SDK users in the next SDK update tomorrow. After some testing, the general feeling is that it represents an incremental improvement rather than a radical leap, although many on social media present it as revolutionary.
Equivalent accuracy in our benchmarks GPT-5 performs similarly to the best existing model, so the improvement in accuracy is, at best, moderate.
Complex tool handling GPT-5 shows a greater willingness to use external tools and seems to prefer broader and more complex tools. This opens up opportunities to build more powerful AI agents, but forces us to rethink the design and selection of the tools used by your solution.
Performance and latency with default parameters the model is significantly slower, between 5 and 10 times in our benchmarks. This makes tuning the reasoning_effort and verbosity parameters key for production projects.
Model control the removal of the model selector forces greater reliance on natural language instructions and the reasoning_effort and verbosity parameters to guide behavior. OpenAI has published prompting guides and optimizers, but there is still work to be done before the model needs less control.
Price GPT-5 consumes slightly more tokens in our tests, approximately between 10 and 20 percent more, but its cost per token is lower than previous versions such as GPT-4o and 4.1, making its price-to-intelligence ratio attractive for many use cases.
Context window with 256k tokens the window is reasonable, although there are cases where 1M token windows, such as those of other models, are necessary for specific projects.
Use in development and code in development tools such as Cursor GPT-5 is slow and sometimes over-analyzes problems, so for iterative and programming assistance tasks we have returned to models such as claude-4, reserving GPT-5 for one-shot or specific tasks.
Promising features there are two new features that we have not yet explored in depth and that may be very useful for agents and advanced workflows
Tool preambles GPT-5 can emit tool preamble messages with progress updates. This helps keep the user informed during the execution of an agent and makes it easier to build more transparent experiences.
Replanning GPT-5 seems to handle scenarios that require rethinking better when a plan fails or enters repetitive loops. It will be interesting to validate whether it reduces common infinite feedback problems.
In summary GPT-5 represents an incremental improvement with interesting strengths and practical limitations to keep in mind. In some cases we still prefer alternative models for programming tasks or highly iterative workflows.
At Q2BSTUDIO as a custom software and application development company we take advantage of these developments to evaluate how to integrate AI agents into enterprise solutions. We are specialists in custom software and custom applications with experience in artificial intelligence and AI for businesses. We also offer cybersecurity services, AWS and Azure cloud services, business intelligence services and solutions with Power BI. We design AI agents to automate processes and improve decision-making, and we combine artificial intelligence capabilities with security consulting and cloud deployment for scalable projects.
If your company needs custom software or wants to explore AI agents, integration of artificial intelligence into products, cybersecurity strategies, migration and deployment on AWS and Azure cloud services, or enhance analysis with Power BI and business intelligence services at Q2BSTUDIO we can help design and implement custom solutions that leverage models like GPT-5 and other emerging technologies.
