My experience with Claude AI over one year
Yesterday, my annual subscription to the American Claude AI service, described as one of the most powerful AI models in the world, came to an end. Of course, the year-long subscription had been taken out by mistake in the first place, and I did not renew it for several reasons, ranging from payment restrictions in my country to matters of priority.
Before talking about the experience, it is important to point out, for those who do not know the writer of these lines, that my perspective comes from being a web developer and a professional in electronic journalism and content management, with more than 18 years of experience. Over the years, I have used artificial intelligence tools in all their forms — from direct chat, to projects and agents, to APIs, and many other tests.
What is my impression?
A year ago, Claude may well have been the best in the world. But over the past year, models have advanced at a rapid pace. OpenAI released stronger versions of ChatGPT, and in my opinion, the paid version of ChatGPT has now become the undisputed leader, capable of handling most tasks with high efficiency.
At the same time, China released massive models, including GLM and Kimi, in addition to DeepSeek. In short, this means that what you hear about Claude and its capabilities is no longer as unique as it once was. Even whatever advantage it still has, if any, no longer makes the kind of difference it did a year ago.
Claude remains one of the best models, but it is also among the most expensive, and it is not available in all countries, unlike ChatGPT and several Chinese models. I had excellent experiences with it and benefited from it greatly, especially in programming. At times, however, its interventions would disrupt my work or force me to redo part of what I had already completed.
After installing GPT Codex and Claude on my computer, I began comparing them in a practical way, and over the past few months, it seemed to me that the balance tipped in favor of GPT in a number of tasks, particularly in terms of quick understanding and retaining a memory that did not require it to search the project from scratch each time. When I asked ChatGPT to carry out a task within one of my projects, it would begin working directly, as if it already had the full picture of the project's structure, its functions, and its original classes. With Claude, however, at the start of almost every session, it would go back and re-check the project, the database, and the core files before proceeding with execution — something that cost me not only extra time, but also an additional portion of my quota, often without any real need for it. That said, this may change if OpenAI imposes restrictions or releases versions that limit usefulness — and this has happened often. A year from now, the landscape may well shift in favor of new rising contenders.
None of this diminishes in any way the brilliance of a great model called Claude AI, or its distinctly human language style. It often apologized in a way that felt like a human apology after making mistakes, and I benefited from it in many tasks. I still consider it a model worthy of respect.
In the end, artificial intelligence is like any other tool: its value depends on the person using it and on their ability to make proper use of it. Outputs depend greatly on the quality of the inputs. The better you know what you want and how to ask for it, the more value you can extract from any model, whatever its name.
