With Gemini 3.1 Pro and Claude Sonnet 4.6 both dropping this week, I wanted to cut through the hype and answer the only question that matters in production: which one is actually better for the price? So I ran three live, real-world tests inside n8n: (1) business “thinking” where the model acts like an MBA/executive, (2) tool calling with real integrations (Perplexity, Google Sheets, Gmail), and (3) information extraction from a dense earnings transcript with both easy fields and math. On the business scenario, Sonnet was way faster (about 1:10 vs Gemini’s 4:14) and it immediately flagged a math discrepancy I accidentally introduced, and it kept calling it out clearly. Gemini did catch the mismatch too, but it was more buried, and the output was shorter (roughly 2.5 pages vs Sonnet’s ~6 pages). On tool calling, both models successfully researched via Perplexity, wrote to Sheets, and emailed me—but Sonnet finished in ~40 seconds and gave more unique tools, while Gemini took over five minutes and leaned on more generic picks. On the extractor test, Sonnet was lightning fast (~9 seconds) but missed one key number, while Gemini was slower (~1:31) and got that enterprise customer count right. My takeaway: Sonnet is the speed + depth play (just set token/output limits), and Gemini can be a solid value when you want tighter outputs and strong extraction accuracy.