Claude Falters as ChatGPT Users Depart, Math Test Reveals Unexpected Gaps

Claude’s recent dip in user engagement is now being quantified alongside a modest but telling performance gap in mathematics, according to a hands‑on test published on Medium. The author, Himansh, pitted Anthropic’s Claude against OpenAI’s ChatGPT on six problems spanning word‑problem reasoning, geometry proofs, basic algebra, statistics and code‑driven verification. Claude “won” on word problems, geometry proofs and the ability to check a user’s work, while ChatGPT’s paid‑tier model outperformed on statistics and any task that required on‑the‑fly Python execution. The two models tied on basic algebra. The test highlighted Claude’s tutor‑like step‑by‑step explanations, which the author says “felt like a tutor, not a calculator,” whereas ChatGPT delivered correct answers but often omitted intermediate reasoning (Medium, Mar 2026).

The same Medium piece also notes a subtle but critical difference in confidence handling. When asked to locate an error in the author’s own solution, Claude identified the mistake, corrected the specific step and openly admitted uncertainty on a borderline portion. By contrast, ChatGPT flagged the error but presented its correction with “high confidence,” even though one component was slightly off. The author warns that such overconfidence can mislead students, a point that underscores why Claude’s more cautious tone may be preferable for educational contexts (Medium, Mar 2026).

These qualitative findings dovetail with a broader usage trend reported by Forbes, which describes Claude as “struggling to cope with ChatGPT exodus.” The article cites a noticeable migration of users from Anthropic’s platform to OpenAI’s, driven in part by the latter’s superior code execution capabilities and the broader ecosystem of plugins and integrations that accompany the paid tier. While Forbes does not provide raw churn numbers, the narrative suggests that the migration is large enough to be noteworthy for Anthropic’s growth outlook (Forbes, Mar 6 2026).

Industry observers have linked the user shift to Anthropic’s recent strategic moves, including the rollout of a revised “Constitution” for Claude aimed at reinforcing ethical behavior. The Verge and TechCrunch both reported on the updated Constitution, noting that Anthropic is positioning the document as a living framework to guide the model’s responses (The Verge; TechCrunch). However, neither outlet connects the constitutional changes to the immediate performance gaps revealed in the math test, nor do they quantify the impact on user retention.

Taken together, the evidence points to a bifurcated market positioning: Claude excels as an explanatory, learning‑oriented assistant, while ChatGPT, especially in its paid tier, dominates computation‑heavy tasks that benefit from real‑time code execution. For enterprises and educators weighing the two, the choice may hinge on whether the priority is transparent reasoning or raw computational throughput. As Anthropic refines Claude’s ethical guardrails, the company will need to address the functional shortfall in code‑driven math to stem the ongoing user migration highlighted by Forbes.