Tencent Hy4 Preview Hits Vercel AI Gateway & OpenRouter

What shipped? Tencent's Hy4 Preview model landed on Vercel AI Gateway and OpenRouter simultaneously — a 770B-parameter MoE behemoth with 49B active parameters and 1M token context window.

What changed? Vercel added Hy4 Preview to AI Gateway on August 28, letting you call it with a single API key — no Tencent Cloud account needed. OpenRouter also added tencent/hy4-preview the same day. Pricing lands at $0.83/M input tokens (prompt) and $2.50/M output (completion), with a half-priced cached-input rate of $0.42/M. That's aggressively cheap for a model this size — roughly the same ballpark as DeepSeek-V3, but aimed at a different niche.

Tencent frames Hy4 as a "coding agent and complex tool-use" model. The architecture is pure MoE text-in/text-out — no vision, no multimodal. What it lacks in modalities it makes up for in scale: 1M context means it can swallow an entire large codebase in one go.

Why a builder cares: Hy4 Preview is the first serious Chinese frontier model to land on Western inference platforms with minimal friction. If you've been running DeepSeek or Qwen for agent loops, Hy4 is now a one-line fallback swap in both Vercel and OpenRouter — no Tencent partnership, no new account. Try it on agentic coding tasks where context length matters and per-token cost adds up.