I Benchmarked 8 Ollama Cloud AI Models. The 397B One Lost to a 1.6s Model.

Speed tests, tool calling, reasoning, and code generation across 8 cloud models. The results surprised me.