Skip to content

Nanbeige4.2 3B VRAM requirements

How much VRAM Nanbeige4.2 3B needs
4-bit weights: 2.4 GBWith 8K context: 3.9 GBWith 32K context: 6.0 GB8-bit, 8K context: 5.7 GB
Parameters
4.2B
Longest context
256K
Released
Jul 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Nanbeige4.2 3B needs

Memory Nanbeige4.2 3B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit7.8 GB9.3 GB11.4 GB19.6 GB
8-bit4.1 GB5.7 GB7.7 GB16.0 GB
6-bit3.2 GB4.7 GB6.8 GB15.0 GB
5-bit2.8 GB4.3 GB6.4 GB14.6 GB
4-bit2.4 GB3.9 GB6.0 GB14.2 GB
3-bit1.9 GB3.4 GB5.5 GB13.8 GB
2-bit1.5 GB3.0 GB5.1 GB13.3 GB
1.6-bit820 MB2.3 GB4.4 GB12.7 GB

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

GPUs and Macs that run Nanbeige4.2 3B

The best way Nanbeige4.2 3B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit28K47 tok/s0.5 sPDF: Nanbeige4.2 3B on RTX 4060 report
RTX 50608-bit28K77 tok/s0.4 sPDF: Nanbeige4.2 3B on RTX 5060 report
RTX 30708-bit28K77 tok/s0.3 sPDF: Nanbeige4.2 3B on RTX 3070 report
RTX 3080 10 GB8-bit52K125 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit72K62 tok/s0.6 sPDF: Nanbeige4.2 3B on RTX 3060 12 GB report
Arc B5808-bit72K54 tok/s0.4 sPDF: Nanbeige4.2 3B on Arc B580 report
RTX 40708-bit72K86 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 4070 report
RTX 4070 Super8-bit72K86 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 4070 Super report
RTX 50708-bit72K112 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit112K50 tok/s0.3 sPDF: Nanbeige4.2 3B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit112K77 tok/s0.3 sPDF: Nanbeige4.2 3B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit112K66 tok/s0.3 sPDF: Nanbeige4.2 3B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit112K89 tok/s0.2 sPDF: Nanbeige4.2 3B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit112K112 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 4070 Ti Super report
RTX 4080 Super8-bit112K122 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX 4080 Super report
RTX 5070 Ti8-bit112K146 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 5070 Ti report
RTX 50808-bit112K155 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX 5080 report
Radeon RX 7900 XTX8-bit200K127 tok/s0.3 sPDF: Nanbeige4.2 3B on Radeon RX 7900 XTX report
RTX 30908-bit200K152 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX 3090 report
RTX 40908-bit200K162 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX 4090 report
Radeon AI PRO R97008-bit256K89 tok/s0.2 sPDF: Nanbeige4.2 3B on Radeon AI PRO R9700 report
RTX 50908-bit256K266 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX 5090 report
2× RTX 30908-bit256K152 tok/s0.1 sPDF: Nanbeige4.2 3B on 2× RTX 3090 report
2× RTX 40908-bit256K162 tok/s0.1 sPDF: Nanbeige4.2 3B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K209 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K155 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX 6000 Ada report
L40S8-bit256K141 tok/s0.1 sPDF: Nanbeige4.2 3B on L40S report
2× RTX 50908-bit256K266 tok/s0.1 sPDF: Nanbeige4.2 3B on 2× RTX 5090 report
4× RTX 30908-bit256K152 tok/s0.1 sPDF: Nanbeige4.2 3B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K266 tok/s0.1 sPDF: Nanbeige4.2 3B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit60K24 tok/s4.5 sPDF: Nanbeige4.2 3B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit120K24 tok/s4.5 sPDF: Nanbeige4.2 3B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit176K24 tok/s4.5 sPDF: Nanbeige4.2 3B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit120K52 tok/s2.3 sPDF: Nanbeige4.2 3B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit256K52 tok/s2.3 sPDF: Nanbeige4.2 3B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K52 tok/s2.3 sPDF: Nanbeige4.2 3B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K89 tok/s1.1 sPDF: Nanbeige4.2 3B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K98 tok/s0.3 sPDF: Nanbeige4.2 3B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K110 tok/s0.6 sPDF: Nanbeige4.2 3B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K110 tok/s0.6 sPDF: Nanbeige4.2 3B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K110 tok/s0.6 sPDF: Nanbeige4.2 3B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K148 tok/s0.2 sPDF: Nanbeige4.2 3B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K38 tok/s0.6 sPDF: Nanbeige4.2 3B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K38 tok/s0.6 sPDF: Nanbeige4.2 3B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K52 tok/s0.2 sPDF: Nanbeige4.2 3B on RTX Spark report
DGX Spark8-bit256K48 tok/s0.2 sPDF: Nanbeige4.2 3B on DGX Spark report
Dual DGX Spark8-bit256K48 tok/s0.1 sPDF: Nanbeige4.2 3B on Dual DGX Spark report
DGX Station GB3008-bit256K693 tok/s0.1 sPDF: Nanbeige4.2 3B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K266 tok/s0.1 sPDF: Nanbeige4.2 3B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K266 tok/s0.1 sPDF: Nanbeige4.2 3B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K266 tok/s0.1 sPDF: Nanbeige4.2 3B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K295 tok/s0.1 sPDF: Nanbeige4.2 3B on A100 80 GB SXM report
H100 SXM8-bit256K431 tok/s0.1 sPDF: Nanbeige4.2 3B on H100 SXM report
4× H200 NVL8-bit256K550 tok/s0.1 sPDF: Nanbeige4.2 3B on 4× H200 NVL report
8× H200 NVL8-bit256K550 tok/s0.1 sPDF: Nanbeige4.2 3B 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/Nanbeige_Nanbeige4.2-3B-GGUF
vLLM
vllm serve Nanbeige/Nanbeige4.2-3B
MLX (Mac)
mlx_lm.chat --model mlx-community/Nanbeige4.2-3B-OptiQ-4bit

Questions

How much VRAM does Nanbeige4.2 3B need?

Nanbeige4.2 3B needs 2.4 GB for its weights at 4-bit, 3.9 GB with an 8K context and 6.0 GB with 32K. At 8-bit, 5.7 GB.

What is the minimum VRAM for Nanbeige4.2 3B?

About 3.4 GB at 3-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 RTX 4060.

How big is Nanbeige4.2 3B?

4.2B parameters: 7.8 GB of weights at 16-bit, 2.4 GB at 4-bit.

Can Nanbeige4.2 3B run on an RTX 4090?

Yes. At 8-bit it holds 200K of context and writes about 162 tokens a second.

More models