Skip to content

Laguna XS 2.1 VRAM requirements

How much VRAM Laguna XS 2.1 needs
4-bit weights: 18.9 GBWith 8K context: 20.0 GBWith 32K context: 20.9 GB8-bit, 8K context: 34.2 GB
Parameters
33.4B
Active per token
3B
Longest context
256K
Released
Jun 2026
License
openmdw-1.1
Strongest at
Not rated

Its licence (openmdw-1.1) is not a standard open-source one: check the terms before commercial use.

How much VRAM Laguna XS 2.1 needs

Memory Laguna XS 2.1 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit62.3 GB63.4 GB64.3 GB68.1 GB
8-bit33.1 GB34.2 GB35.1 GB38.9 GB
6-bit25.5 GB26.6 GB27.6 GB31.3 GB
5-bit22.2 GB23.2 GB24.2 GB27.9 GB
4-bit18.9 GB20.0 GB20.9 GB24.6 GB
3-bit15.2 GB16.3 GB17.2 GB21.0 GB
2-bit11.7 GB12.8 GB13.7 GB17.4 GB
1.6-bit6.4 GB7.5 GB8.4 GB12.2 GB

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

GPUs and Macs that run Laguna XS 2.1

The best way Laguna XS 2.1 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Laguna XS 2.1 on RTX 4060 report
RTX 5060Too bigPDF: Laguna XS 2.1 on RTX 5060 report
RTX 3070Too bigPDF: Laguna XS 2.1 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Laguna XS 2.1 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Laguna XS 2.1 on RTX 3060 12 GB report
Arc B580Too bigPDF: Laguna XS 2.1 on Arc B580 report
RTX 4070Too bigPDF: Laguna XS 2.1 on RTX 4070 report
RTX 4070 SuperToo bigPDF: Laguna XS 2.1 on RTX 4070 Super report
RTX 5070Too bigPDF: Laguna XS 2.1 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Laguna XS 2.1 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Laguna XS 2.1 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Laguna XS 2.1 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Laguna XS 2.1 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Laguna XS 2.1 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Laguna XS 2.1 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Laguna XS 2.1 on RTX 5070 Ti report
RTX 5080Too bigPDF: Laguna XS 2.1 on RTX 5080 report
Radeon RX 7900 XTX4-bit80K148 tok/s0.9 sPDF: Laguna XS 2.1 on Radeon RX 7900 XTX report
RTX 30904-bit80K193 tok/s0.5 sPDF: Laguna XS 2.1 on RTX 3090 report
RTX 40904-bit80K200 tok/s0.2 sPDF: Laguna XS 2.1 on RTX 4090 report
Radeon AI PRO R97006-bit104K100 tok/s0.6 sPDF: Laguna XS 2.1 on Radeon AI PRO R9700 report
RTX 50906-bit104K229 tok/s0.2 sPDF: Laguna XS 2.1 on RTX 5090 report
2× RTX 30908-bit256K137 tok/s0.3 sPDF: Laguna XS 2.1 on 2× RTX 3090 report
2× RTX 40908-bit256K144 tok/s0.1 sPDF: Laguna XS 2.1 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K172 tok/s0.3 sPDF: Laguna XS 2.1 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K139 tok/s0.2 sPDF: Laguna XS 2.1 on RTX 6000 Ada report
L40S8-bit256K130 tok/s0.3 sPDF: Laguna XS 2.1 on L40S report
2× RTX 50908-bit256K202 tok/s0.1 sPDF: Laguna XS 2.1 on 2× RTX 5090 report
4× RTX 30908-bit256K137 tok/s0.2 sPDF: Laguna XS 2.1 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K202 tok/s0.1 sPDF: Laguna XS 2.1 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Laguna XS 2.1 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Laguna XS 2.1 on Mac mini M4 24 GB report
Mac mini M4 32 GB4-bit20K41 tok/s12 sPDF: Laguna XS 2.1 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Laguna XS 2.1 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB6-bit200K60 tok/s6.2 sPDF: Laguna XS 2.1 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K50 tok/s6.2 sPDF: Laguna XS 2.1 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K74 tok/s3.1 sPDF: Laguna XS 2.1 on M4 Max 128 GB report
M5 Max 128 GB8-bit256K80 tok/s0.9 sPDF: Laguna XS 2.1 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K86 tok/s1.7 sPDF: Laguna XS 2.1 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K86 tok/s1.7 sPDF: Laguna XS 2.1 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K86 tok/s1.7 sPDF: Laguna XS 2.1 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K103 tok/s0.5 sPDF: Laguna XS 2.1 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K41 tok/s1.8 sPDF: Laguna XS 2.1 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K41 tok/s1.8 sPDF: Laguna XS 2.1 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K57 tok/s0.7 sPDF: Laguna XS 2.1 on RTX Spark report
DGX Spark8-bit256K52 tok/s0.7 sPDF: Laguna XS 2.1 on DGX Spark report
Dual DGX Spark8-bit256K52 tok/s0.4 sPDF: Laguna XS 2.1 on Dual DGX Spark report
DGX Station GB3008-bit256K328 tok/s0.1 sPDF: Laguna XS 2.1 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K202 tok/s0.1 sPDF: Laguna XS 2.1 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K202 tok/s0.1 sPDF: Laguna XS 2.1 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K202 tok/s0.1 sPDF: Laguna XS 2.1 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K215 tok/s0.4 sPDF: Laguna XS 2.1 on A100 80 GB SXM report
H100 SXM8-bit256K265 tok/s0.1 sPDF: Laguna XS 2.1 on H100 SXM report
4× H200 NVL8-bit256K298 tok/s0.1 sPDF: Laguna XS 2.1 on 4× H200 NVL report
8× H200 NVL8-bit256K298 tok/s0.1 sPDF: Laguna XS 2.1 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 mudler/Laguna-XS-2.1-APEX-GGUF
vLLM
vllm serve poolside/Laguna-XS-2.1
MLX (Mac)
mlx_lm.chat --model mlx-community/Laguna-XS-2.1-4bit

Questions

How much VRAM does Laguna XS 2.1 need?

Laguna XS 2.1 needs 18.9 GB for its weights at 4-bit, 20.0 GB with an 8K context and 20.9 GB with 32K. At 8-bit, 34.2 GB.

What is the minimum VRAM for Laguna XS 2.1?

About 16.3 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 Radeon RX 7900 XTX.

How big is Laguna XS 2.1?

33.4B parameters (3B active per token): 62.3 GB of weights at 16-bit, 18.9 GB at 4-bit.

Can Laguna XS 2.1 run on an RTX 4090?

Yes. At 4-bit it holds 80K of context and writes about 200 tokens a second.

More models