- Sources: primary, discussion
- Summary: Tencent released and open-sourced Hy4 preview, describing a 770B-parameter model with 49B active parameters and a context window over 1M tokens. The announcement prices it at 0.834 US dollars per million input tokens, 2.501 US dollars per million output tokens, and 0.042 US dollars per million tokens on cache hits. It also reports a score of 2.99 out of 4.00 against GLM-5.3 and Kimi K3. That number comes from an evaluation Tencent ran internally with its own experts, which is a disclosed method and not a reproducible one, so this block rests on the architecture, the context length and the pricing instead.
- Why it matters: An open-weight model at this size and context length adds a self-hostable option at the top of the open model tier.
send feedback on this story