Claude Sonnet 5 vs Hy3

The all-rounder in the Claude family, and a sensible default for most work. Sonnet 5 codes, writes, and analyses across the same ground as the flagship, reads images and documents, and holds the same very large context. It is faster and costs less. It gives up ground only on problems difficult enough to need the flagship.

Tencent’s open-weight reasoning model, built for agents rather than chat. Hy3 blends fast and slow thinking, switching between quick replies and deep reasoning, and Tencent tuned it hard for real production work: tool use, long agent loops, and search. It activates only a fraction of its parameters per token, so it is unusually cheap to run for its capability. Text only. Open weights, Apache 2.0.