FitMyLLM
Tencent/Mixture of Experts

THunyuan Hy4-preview

Hunyuan Hy4-preview is Tencent's 770B MoE flagship preview with 49B active, gated DeepSeek-style sparse attention over MLA, and a 1M context. Tencent says it ships with known issues, such as over-long reasoning, and calls it an early version of Hy4.

chatcodingreasoningmultilingualmathagentictool_use
779.96B
Parameters (49B active)
1024K
Context length
2
Benchmarks
17
Quantizations
12K
HF downloads
Architecture
MoE
Released
2026-08-28
Layers
78
KV Heads
8
Head Dim
64
Family
hunyuan

Quantization Options

Context length:
QuantBitsVRAM @ 16KQuality
IQ2_XXS2.38
233.6 GB
232.5 + 1.0 KV
low
IQ2_M2.93
287.2 GB
286.1 + 1.0 KV
low
Q2_K3.16
309.6 GB
308.6 + 1.0 KV
low
IQ3_XXS3.25
318.4 GB
317.3 + 1.0 KV
low
IQ3_XS3.5
342.7 GB
341.7 + 1.0 KV
low
Q3_K_S3.64
356.4 GB
355.4 + 1.0 KV
low
IQ3_M3.76
368.1 GB
367.1 + 1.0 KV
low
Q3_K_M4
391.5 GB
390.5 + 1.0 KV
low
Q3_K_L4.3
420.7 GB
419.7 + 1.0 KV
moderate
IQ4_XS4.46
436.3 GB
435.3 + 1.0 KV
moderate
Q4_K_S4.67
456.8 GB
455.8 + 1.0 KV
moderate
Q4_K_M4.89
478.3 GB
477.2 + 1.0 KV
good
Q5_K_S5.57
544.6 GB
543.5 + 1.0 KV
good
Q5_K_M5.7
557.2 GB
556.2 + 1.0 KV
good
Q6_K6.56
641.1 GB
640.1 + 1.0 KV
excellent
Q8_08.5
830.2 GB
829.2 + 1.0 KV
lossless
FP1616
1561.4 GB
1560.4 + 1.0 KV
lossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Too big for a single GPU — plan a multi-GPU deployment
Even the lightest quant needs ~234 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →
▸ READY TO RUN THIS?RENT BY THE HOUR

RENT A GPU AND RUN HUNYUAN HY4-PREVIEW NOW

Rent on Vast.ai →Or RunPod →

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (2)

GPQA Diamond92.3
HLE43.4

Run this model

▸Easiest way to get started·Beginners
DOCS ↗
curl -fsSL https://ollama.com/install.sh | sh
$ollama run hunyuan:780b-q4_K_M

Tag may need adjustment — check ollama.com/library/hunyuan for available tags.

▸ SETUP GUIDE
>_

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more
Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines
HuggingFaceGGUF DownloadsBuild Hardware

Find the best GPU for Hunyuan Hy4-preview

Build Hardware for Hunyuan Hy4-preview
▸ SPEC SHEET

Hunyuan Hy4-preview — 779.96B MoE.

▸ SPECIFICATIONS
PARAMETERS
779.96B (49B active)
ARCHITECTURE
Mixture of Experts
CONTEXT LENGTH
1024K tokens
CAPABILITIES
chat, coding, reasoning, multilingual, math, agentic, tool_use
RELEASE DATE
2026-08-28
PROVIDER
Tencent
FAMILY
hunyuan
▸ VRAM REQUIREMENTS
QUANTBPWVRAMQUALITY
IQ2_XXS2.38232.5 GB65%
IQ2_M2.93286.1 GB75%
Q2_K3.16308.6 GB78%
IQ3_XXS3.25317.3 GB82%
IQ3_XS3.5341.7 GB84%
Q3_K_S3.64355.4 GB85%
IQ3_M3.76367.1 GB86%
Q3_K_M4390.5 GB88%
Q3_K_L4.3419.7 GB90%
IQ4_XS4.46435.3 GB92%
Q4_K_S4.67455.8 GB93%
Q4_K_M4.89477.2 GB94%
Q5_K_S5.57543.5 GB96%
Q5_K_M5.7556.2 GB96%
Q6_K6.56640.1 GB97%
Q8_08.5829.2 GB100%
FP16161560.4 GB100%
§ 01BENCHMARK SCORES
GPQA Diamond92.3
HLE43.4