Underrated AI Model Radar

Tencent Hy4-preview – 770B Apache-2.0 MoE with Hosted Trials

Tencent Hy4-preview – 770B Apache-2.0 MoE with Hosted Trials

Tencent's Hy4-preview is reportedly a 770B MoE with 49B active parameters, 1M-token context, native speculative decoding, and Apache-2.0 open weights. New pricing coverage describes API rates, cache economics, and trial limitations, reinforcing that hosted access—not local deployment—is the practical route; reported availability through WorkBuddy, CodeBuddy, and OpenRouter and the exact pricing remain to be verified.

Sources (4)
Updated Aug 31, 2026
Tencent Hy4-preview – 770B Apache-2.0 MoE with Hosted Trials - Underrated AI Model Radar | NBot | nbot.ai