Skip to content

LLaDA2.1 Mini VRAM requirements

How much VRAM LLaDA2.1 Mini needs
4-bit weights: 9.2 GBWith 8K context: 10.3 GBWith 32K context: 11.2 GB8-bit, 8K context: 17.2 GB
Parameters
16.3B
Active per token
1.4B
Longest context
32K
Released
Feb 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM LLaDA2.1 Mini needs

Memory LLaDA2.1 Mini needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit30.3 GB31.4 GB32.3 GB
8-bit16.1 GB17.2 GB18.1 GB
6-bit12.4 GB13.5 GB14.5 GB
5-bit10.8 GB11.9 GB12.8 GB
4-bit9.2 GB10.3 GB11.2 GB
3-bit7.4 GB8.5 GB9.5 GB
2-bit5.7 GB6.8 GB7.7 GB
1.6-bit3.1 GB4.2 GB5.2 GB

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

GPUs and Macs that run LLaDA2.1 Mini

The best way LLaDA2.1 Mini runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: LLaDA2.1 Mini on RTX 4060 report
RTX 5060Too bigPDF: LLaDA2.1 Mini on RTX 5060 report
RTX 3070Too bigPDF: LLaDA2.1 Mini on RTX 3070 report
RTX 3080 10 GB3-bit32K397 tok/s0.2 sPDF: LLaDA2.1 Mini on RTX 3080 10 GB report
RTX 3060 12 GB4-bit32K215 tok/s0.5 sPDF: LLaDA2.1 Mini on RTX 3060 12 GB report
Arc B5804-bit32K172 tok/s0.3 sPDF: LLaDA2.1 Mini on Arc B580 report
RTX 40704-bit32K273 tok/s0.2 sPDF: LLaDA2.1 Mini on RTX 4070 report
RTX 4070 Super4-bit32K273 tok/s0.2 sPDF: LLaDA2.1 Mini on RTX 4070 Super report
RTX 50704-bit32K328 tok/s0.2 sPDF: LLaDA2.1 Mini on RTX 5070 report
RTX 4060 Ti 16 GB6-bit32K142 tok/s0.3 sPDF: LLaDA2.1 Mini on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB6-bit32K202 tok/s0.3 sPDF: LLaDA2.1 Mini on RTX 5060 Ti 16 GB report
Arc A770 16 GB6-bit32K161 tok/s0.3 sPDF: LLaDA2.1 Mini on Arc A770 16 GB report
Radeon RX 9070 XT6-bit32K206 tok/s0.2 sPDF: LLaDA2.1 Mini on Radeon RX 9070 XT report
RTX 4070 Ti Super6-bit32K270 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 4070 Ti Super report
RTX 4080 Super6-bit32K287 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 4080 Super report
RTX 5070 Ti6-bit32K325 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 5070 Ti report
RTX 50806-bit32K339 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 5080 report
Radeon RX 7900 XTX8-bit32K225 tok/s0.3 sPDF: LLaDA2.1 Mini on Radeon RX 7900 XTX report
RTX 30908-bit32K284 tok/s0.2 sPDF: LLaDA2.1 Mini on RTX 3090 report
RTX 40908-bit32K298 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 4090 report
Radeon AI PRO R97008-bit32K173 tok/s0.2 sPDF: LLaDA2.1 Mini on Radeon AI PRO R9700 report
RTX 50908-bit32K414 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 5090 report
2× RTX 30908-bit32K284 tok/s0.1 sPDF: LLaDA2.1 Mini on 2× RTX 3090 report
2× RTX 40908-bit32K298 tok/s0.1 sPDF: LLaDA2.1 Mini on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K355 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K289 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX 6000 Ada report
L40S8-bit32K269 tok/s0.1 sPDF: LLaDA2.1 Mini on L40S report
2× RTX 50908-bit32K414 tok/s0.1 sPDF: LLaDA2.1 Mini on 2× RTX 5090 report
4× RTX 30908-bit32K284 tok/s0.1 sPDF: LLaDA2.1 Mini on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K414 tok/s0.1 sPDF: LLaDA2.1 Mini on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB3-bit32K101 tok/s4.0 sPDF: LLaDA2.1 Mini on Mac mini M4 16 GB report
Mac mini M4 24 GB6-bit32K68 tok/s4.0 sPDF: LLaDA2.1 Mini on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K55 tok/s4.0 sPDF: LLaDA2.1 Mini on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB6-bit32K124 tok/s2.0 sPDF: LLaDA2.1 Mini on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K104 tok/s2.0 sPDF: LLaDA2.1 Mini on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K104 tok/s2.0 sPDF: LLaDA2.1 Mini on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K153 tok/s1.0 sPDF: LLaDA2.1 Mini on M4 Max 128 GB report
M5 Max 128 GB8-bit32K164 tok/s0.3 sPDF: LLaDA2.1 Mini on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K177 tok/s0.6 sPDF: LLaDA2.1 Mini on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K177 tok/s0.6 sPDF: LLaDA2.1 Mini on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K177 tok/s0.6 sPDF: LLaDA2.1 Mini on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K210 tok/s0.2 sPDF: LLaDA2.1 Mini on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K85 tok/s0.6 sPDF: LLaDA2.1 Mini on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K85 tok/s0.6 sPDF: LLaDA2.1 Mini on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K118 tok/s0.3 sPDF: LLaDA2.1 Mini on RTX Spark report
DGX Spark8-bit32K109 tok/s0.4 sPDF: LLaDA2.1 Mini on DGX Spark report
Dual DGX Spark8-bit32K109 tok/s0.2 sPDF: LLaDA2.1 Mini on Dual DGX Spark report
DGX Station GB3008-bit32K664 tok/s0.1 sPDF: LLaDA2.1 Mini on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K414 tok/s0.1 sPDF: LLaDA2.1 Mini on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K414 tok/s0.1 sPDF: LLaDA2.1 Mini on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K414 tok/s0.1 sPDF: LLaDA2.1 Mini on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K441 tok/s0.1 sPDF: LLaDA2.1 Mini on A100 80 GB SXM report
H100 SXM8-bit32K541 tok/s0.1 sPDF: LLaDA2.1 Mini on H100 SXM report
4× H200 NVL8-bit32K605 tok/s0.1 sPDF: LLaDA2.1 Mini on 4× H200 NVL report
8× H200 NVL8-bit32K605 tok/s0.1 sPDF: LLaDA2.1 Mini 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

vLLM
vllm serve inclusionAI/LLaDA2.1-mini
MLX (Mac)
mlx_lm.chat --model mlx-community/LLaDA2.1-mini-4bit

Questions

How much VRAM does LLaDA2.1 Mini need?

LLaDA2.1 Mini needs 9.2 GB for its weights at 4-bit, 10.3 GB with an 8K context and 11.2 GB with 32K. At 8-bit, 17.2 GB.

What is the minimum VRAM for LLaDA2.1 Mini?

About 8.5 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 3080 10 GB.

How big is LLaDA2.1 Mini?

16.3B parameters (1.4B active per token): 30.3 GB of weights at 16-bit, 9.2 GB at 4-bit.

Can LLaDA2.1 Mini run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 298 tokens a second.

More models