Size
770B
49B active — only part of the model works on each answer, but all of it has to fit in memory
Hy4
Context
Not yet confirmed
the longest conversation the model is built to hold
Lab
Tencent — Hunyuan
the lab that released it
Released
Nov 4, 2026
Licence: Not yet stated — Tencent says the weights and system card will be released in the coming days; no licence published yet.
Worth knowing
Tencent's flagship Hunyuan model, released Nov 4, 2026: a 770B-parameter MoE (49B active, in Tencent's launch figures) with vision and computer use for the first time in the line. Live today on Tencent Cloud TokenHub and in Tencent's apps at about $0.90 in / $2.70 out per 1M tokens (Tencent's own release); the open weights are promised in the coming days, so there is nothing to self-host yet. At 770B parameters even the 4-bit weights alone would be about 385GB — a multi-GPU job when they land.
What it needs
Full precision (FP16)
1,540 GB
the weights, uncompressed
8-bit
770 GB
the weights, compressed to 8-bit
4-bit
385 GB
the weights, compressed to 4-bit
How we work this out: memory = the model's total parameters × bytes per weight — 2 bytes at full precision (FP16), 1 byte at 8-bit, half a byte at 4-bit. Weights only — add room for the conversation memory (KV cache) and the serving software. A GPU counts as running a build when its memory covers the weights plus about 20% headroom. Real figures move with context length and software.
Weights not yet released
Hy4’s hosted API is live, but its downloadable weights have not been released yet — nobody can self-host it for now. The GPU ranking for it appears here the day the weights land.