Habr AI pits Claude, Gemini and ChatGPT against each other on writing, math, analysis and creativity

Habr AI has published a comparison of three flagship models: ChatGPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro. Instead of the usual "write a story" or "solve this problem" tests, the author shifted the focus to unconventional scenarios, where the real differences between the systems are easier to see.

An unconventional test format

The main idea of the piece is not to find an absolute winner, but to check how the models behave outside the most overused demos. When LLMs are asked to write a short text, generate a code template or solve a school-level problem, they often perform at a similar level. But in stranger, borderline or simply less formulaic tasks, differences begin to show: style of thinking, flexibility, resilience to ambiguity, and the ability to hold context without hints at every step.

This approach also matters because users increasingly apply models not for a single isolated command but as an intellectual tool for work. In real practice, it is not enough to "answer correctly". The model also has to grasp the hidden requirement, not stumble on the wording, not drift into needless chatter and not lose the logic midway through its reasoning. That is why a comparison built on unusual tasks looks more useful than yet another formal benchmark.

Three flagship models

The test features ChatGPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro, three systems that usually dominate discussions of generation quality. The lineup alone shows this is not a niche experiment but a comparison of current flagships, the ones advanced users, editors, analysts and teams with LLMs built into daily workflows most often choose between.

The author also does not try to present the piece as a final verdict on the market. It is rather an attempt to answer a more practical question: where exactly do the differences between the best models become noticeable. On ordinary tasks the gap may be small, but in scenarios with ambiguity, combined requirements and creative constraints, each model shows its own signature. For the reader this is more useful than a dry ranking, because it helps match a model to a specific type of work.

What the test looks at

According to the article, the focus is on four groups of tasks that are closer to real use than to showcase presentations. The aim is not to check a single metric but to see how a model switches between different types of thinking: from careful editing and formal logic to calculation and free-form idea generation. This set evaluates not one narrow skill but the system's behavior in different modes, from precision to creativity.

• Working with text and the quality of phrasing
• Math and resistance to calculation errors
• Analytical tasks with several conditions
• Creativity in atypical and not fully formalized requests

The strength of this kind of comparison is that it shows not only a model's level of knowledge but also the character of its answers. One system may be neater in structure, another bolder in ideas, a third more stable in logic. For the user this often matters more than an abstract first place, because the choice of LLM depends not on general hype but on what exactly needs doing: editing text, checking reasoning, solving problems or quickly finding unconventional approaches.

What this means

Comparisons like this are gradually changing how we talk about LLMs. The question is no longer "who is smarter overall" but which model handles your real scenario best. For the market this is a sign of maturity: flagships have become strong enough to be judged not by wow factor but by the nuances of how they work.