Everyone’s switching from ChatGPT to Claude — but new tests say neither is the smartest free AI, and the real winner might surprise you
A recent report from OmniCalculator evaluates leading AI chatbots and finds that while Claude is favored for writing quality and tone, and ChatGPT remains the most popular, neither is the top performer in logic and problem-solving. xAI's Grok 4.2 outperforms both in mathematical reasoning and consistency during complex tasks, despite being less polished in style. The findings suggest that perceived intelligence in AI may not always align with actual performance on technical benchmarks.
- ▪OmniCalculator's testing indicates that Grok 4.2 excels in logic and problem-solving compared to other free AI models.
- ▪Claude 4.6 is rated highest for writing quality, coherence, and maintaining tone across long responses.
- ▪ChatGPT remains the most widely used AI chatbot despite a growing shift toward Claude.
- ▪Grok 4.2 shows significantly lower answer instability, revising its responses only 33.1% of the time in complex scenarios.
- ▪Legacy models like earlier versions of ChatGPT and Claude revise their answers about 60% of the time during complex reasoning tasks.
Opening excerpt (first ~120 words) tap to expand
AI Platforms & Assistants OpenAI ChatGPT Everyone’s switching from ChatGPT to Claude — but new tests say neither is the smartest free AI, and the real winner might surprise you News By Eric Hal Schwartz published 1 May 2026 Claude may feel smarter but the data tells a different story When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works. (Image credit: Getty Images) Copy link Facebook X Whatsapp Reddit Pinterest Flipboard Threads Email Share this article 0 Join the conversation Follow us Add us as a preferred source on Google Newsletter Subscribe to our newsletter Testing from OmniCalculator suggests Claude and ChatGPT are not the smartestThe report finds Grok 4.2 performs best in logic and problem-solvingClaude still leads in writing…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Latest from TechRadar .
Discussion
0 commentsMore from Latest from TechRadar
