Skip to content

Hy3 VRAM requirements

How much VRAM Hy3 needs
4-bit weights: 167 GBWith 8K context: 170 GBWith 32K context: 177 GB8-bit, 8K context: 295 GB
Parameters
295B
Active per token
21B
Longest context
256K
Released
Jul 2026
License
Apache 2.0
Strongest at
Chat, Coding, Agents, Docs

How much VRAM Hy3 needs

Memory Hy3 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit549 GB553 GB560 GB590 GB
8-bit292 GB295 GB303 GB333 GB
6-bit225 GB229 GB236 GB266 GB
5-bit195 GB199 GB206 GB236 GB
4-bit167 GB170 GB177 GB207 GB
3-bit134 GB138 GB145 GB175 GB
2-bit103 GB106 GB114 GB144 GB
1.6-bit56.7 GB60.0 GB67.5 GB97.5 GB

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

GPUs and Macs that run Hy3

The best way Hy3 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Hy3 on RTX 4060 report
RTX 5060Too bigPDF: Hy3 on RTX 5060 report
RTX 3070Too bigPDF: Hy3 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Hy3 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Hy3 on RTX 3060 12 GB report
Arc B580Too bigPDF: Hy3 on Arc B580 report
RTX 4070Too bigPDF: Hy3 on RTX 4070 report
RTX 4070 SuperToo bigPDF: Hy3 on RTX 4070 Super report
RTX 5070Too bigPDF: Hy3 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Hy3 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Hy3 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Hy3 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Hy3 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Hy3 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Hy3 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Hy3 on RTX 5070 Ti report
RTX 5080Too bigPDF: Hy3 on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Hy3 on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Hy3 on RTX 3090 report
RTX 4090Too bigPDF: Hy3 on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Hy3 on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Hy3 on RTX 5090 report
2× RTX 3090Too bigPDF: Hy3 on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Hy3 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Hy3 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Hy3 on RTX 6000 Ada report
L40SToo bigPDF: Hy3 on L40S report
2× RTX 5090Too bigPDF: Hy3 on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Hy3 on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Hy3 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Hy3 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Hy3 on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Hy3 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Hy3 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Hy3 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Hy3 on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Hy3 on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Hy3 on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Hy3 on M3 Ultra 96 GB report
M3 Ultra 256 GB4-bit48K28 tok/s13 sPDF: Hy3 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit224K19 tok/s13 sPDF: Hy3 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit224K25 tok/s4.1 sPDF: Hy3 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Hy3 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Hy3 on Ryzen AI Max+ 395 128 GB report
RTX Spark2-bit32K24 tok/s4.8 sPDF: Hy3 on RTX Spark report
DGX Spark2-bit32K22 tok/s4.8 sPDF: Hy3 on DGX Spark report
Dual DGX Spark5-bit96K12 tok/s2.7 sPDF: Hy3 on Dual DGX Spark report
DGX Station GB3006-bit40K121 tok/s0.4 sPDF: Hy3 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell4-bit44K67 tok/s0.6 sPDF: Hy3 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit224K44 tok/s0.3 sPDF: Hy3 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K44 tok/s0.2 sPDF: Hy3 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Hy3 on A100 80 GB SXM report
H100 SXMToo bigPDF: Hy3 on H100 SXM report
4× H200 NVL8-bit256K87 tok/s0.3 sPDF: Hy3 on 4× H200 NVL report
8× H200 NVL8-bit256K87 tok/s0.2 sPDF: Hy3 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 bartowski/Hy3-GGUF
vLLM
vllm serve tencent/Hy3
MLX (Mac)
mlx_lm.chat --model mlx-community/Hy3-OptiQ-2bit

Questions

How much VRAM does Hy3 need?

Hy3 needs 167 GB for its weights at 4-bit, 170 GB with an 8K context and 177 GB with 32K. At 8-bit, 295 GB.

What is the minimum VRAM for Hy3?

About 106 GB at 2-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 Hy3?

295B parameters (21B active per token): 549 GB of weights at 16-bit, 167 GB at 4-bit.

Can Hy3 run on an RTX 4090?

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

More models