Skip to content

Hy4 Preview VRAM requirements

How much VRAM Hy4 Preview needs
4-bit weights: 440 GBWith 8K context: 442 GBWith 32K context: 444 GB8-bit, 8K context: 773 GB
Parameters
780B
Active per token
49.7B
Longest context
1M
Released
Aug 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Hy4 Preview needs

Memory Hy4 Preview needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit1,453 GB1,454 GB1,457 GB1,466 GB
8-bit772 GB773 GB776 GB785 GB
6-bit596 GB597 GB600 GB609 GB
5-bit517 GB518 GB521 GB530 GB
4-bit440 GB442 GB444 GB453 GB
3-bit355 GB357 GB359 GB368 GB
2-bit272 GB274 GB276 GB285 GB
1.6-bit150 GB151 GB154 GB163 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Hy4 Preview

The best way Hy4 Preview runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Hy4 Preview on RTX 4060 report
RTX 5060Too bigPDF: Hy4 Preview on RTX 5060 report
RTX 3070Too bigPDF: Hy4 Preview on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Hy4 Preview on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Hy4 Preview on RTX 3060 12 GB report
Arc B580Too bigPDF: Hy4 Preview on Arc B580 report
RTX 4070Too bigPDF: Hy4 Preview on RTX 4070 report
RTX 4070 SuperToo bigPDF: Hy4 Preview on RTX 4070 Super report
RTX 5070Too bigPDF: Hy4 Preview on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Hy4 Preview on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Hy4 Preview on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Hy4 Preview on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Hy4 Preview on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Hy4 Preview on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Hy4 Preview on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Hy4 Preview on RTX 5070 Ti report
RTX 5080Too bigPDF: Hy4 Preview on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Hy4 Preview on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Hy4 Preview on RTX 3090 report
RTX 4090Too bigPDF: Hy4 Preview on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Hy4 Preview on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Hy4 Preview on RTX 5090 report
2× RTX 3090Too bigPDF: Hy4 Preview on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Hy4 Preview on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Hy4 Preview on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Hy4 Preview on RTX 6000 Ada report
L40SToo bigPDF: Hy4 Preview on L40S report
2× RTX 5090Too bigPDF: Hy4 Preview on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Hy4 Preview on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Hy4 Preview on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Hy4 Preview on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Hy4 Preview on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Hy4 Preview on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Hy4 Preview on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Hy4 Preview on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Hy4 Preview on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Hy4 Preview on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Hy4 Preview on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Hy4 Preview on M3 Ultra 96 GB report
M3 Ultra 256 GB1.6-bit320K33 tok/s32 sPDF: Hy4 Preview on M3 Ultra 256 GB report
M3 Ultra 512 GB3-bit88K18 tok/s32 sPDF: Hy4 Preview on M3 Ultra 512 GB report
M5 Ultra 512 GB3-bit88K24 tok/s9.9 sPDF: Hy4 Preview on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Hy4 Preview on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Hy4 Preview on Ryzen AI Max+ 395 128 GB report
RTX SparkToo bigPDF: Hy4 Preview on RTX Spark report
DGX SparkToo bigPDF: Hy4 Preview on DGX Spark report
Dual DGX Spark1.6-bit800K17 tok/s6.5 sPDF: Hy4 Preview on Dual DGX Spark report
DGX Station GB3001.6-bit928K148 tok/s0.9 sPDF: Hy4 Preview on DGX Station GB300 report
2× RTX PRO 6000 Blackwell1.6-bit320K78 tok/s1.5 sPDF: Hy4 Preview on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell3-bit72K42 tok/s0.8 sPDF: Hy4 Preview on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell6-bit1M27 tok/s0.4 sPDF: Hy4 Preview on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Hy4 Preview on A100 80 GB SXM report
H100 SXMToo bigPDF: Hy4 Preview on H100 SXM report
4× H200 NVL5-bit168K66 tok/s0.8 sPDF: Hy4 Preview on 4× H200 NVL report
8× H200 NVL8-bit1M49 tok/s0.4 sPDF: Hy4 Preview on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf 6block/Hy4-preview-GGUF
vLLM
vllm serve tencent/Hy4-preview
MLX (Mac)
mlx_lm.chat --model mlx-community/Hy4-preview-4bit

Questions

How much VRAM does Hy4 Preview need?

Hy4 Preview needs 440 GB for its weights at 4-bit, 442 GB with an 8K context and 444 GB with 32K. At 8-bit, 773 GB.

What is the minimum VRAM for Hy4 Preview?

About 151 GB at 1.6-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the M3 Ultra 256 GB.

How big is Hy4 Preview?

780B parameters (49.7B active per token): 1,453 GB of weights at 16-bit, 440 GB at 4-bit.

Can Hy4 Preview run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the M3 Ultra 256 GB.

More models