Skip to content

LightOnOCR 2 1B VRAM requirements

How much VRAM LightOnOCR 2 1B needs
4-bit weights: 581 MBWith 8K context: 2.2 GBWith 16K context: 3.1 GB8-bit, 8K context: 2.6 GB
Parameters
1B
Longest context
16K
Released
Jan 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM LightOnOCR 2 1B needs

Memory LightOnOCR 2 1B needs at each precision and context length
PrecisionWeightsWith 8KWith 16K
16-bit1.9 GB3.5 GB4.4 GB
8-bit1.0 GB2.6 GB3.5 GB
6-bit786 MB2.4 GB3.3 GB
5-bit682 MB2.3 GB3.2 GB
4-bit581 MB2.2 GB3.1 GB
3-bit469 MB2.1 GB3.0 GB
2-bit360 MB2.0 GB2.9 GB
1.6-bit198 MB1.8 GB2.7 GB

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

GPUs and Macs that run LightOnOCR 2 1B

The best way LightOnOCR 2 1B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit16K155 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4060 report
RTX 50608-bit16K235 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 5060 report
RTX 30708-bit16K235 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 3070 report
RTX 3080 10 GB8-bit16K351 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit16K197 tok/s0.2 sPDF: LightOnOCR 2 1B on RTX 3060 12 GB report
Arc B5808-bit16K157 tok/s0.1 sPDF: LightOnOCR 2 1B on Arc B580 report
RTX 40708-bit16K258 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4070 report
RTX 4070 Super8-bit16K258 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4070 Super report
RTX 50708-bit16K321 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit16K163 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit16K235 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit16K182 tok/s0.1 sPDF: LightOnOCR 2 1B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit16K239 tok/s0.1 sPDF: LightOnOCR 2 1B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit16K321 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4070 Ti Super report
RTX 4080 Super8-bit16K343 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4080 Super report
RTX 5070 Ti8-bit16K393 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 5070 Ti report
RTX 50808-bit16K412 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 5080 report
Radeon RX 7900 XTX8-bit16K307 tok/s0.1 sPDF: LightOnOCR 2 1B on Radeon RX 7900 XTX report
RTX 30908-bit16K405 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 3090 report
RTX 40908-bit16K425 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 4090 report
Radeon AI PRO R97008-bit16K239 tok/s0.1 sPDF: LightOnOCR 2 1B on Radeon AI PRO R9700 report
RTX 50908-bit16K591 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 5090 report
2× RTX 30908-bit16K405 tok/s0.1 sPDF: LightOnOCR 2 1B on 2× RTX 3090 report
2× RTX 40908-bit16K425 tok/s0.1 sPDF: LightOnOCR 2 1B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit16K506 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit16K412 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX 6000 Ada report
L40S8-bit16K384 tok/s0.1 sPDF: LightOnOCR 2 1B on L40S report
2× RTX 50908-bit16K591 tok/s0.1 sPDF: LightOnOCR 2 1B on 2× RTX 5090 report
4× RTX 30908-bit16K405 tok/s0.1 sPDF: LightOnOCR 2 1B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit16K591 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit16K76 tok/s1.2 sPDF: LightOnOCR 2 1B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit16K76 tok/s1.2 sPDF: LightOnOCR 2 1B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit16K76 tok/s1.2 sPDF: LightOnOCR 2 1B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit16K143 tok/s0.6 sPDF: LightOnOCR 2 1B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit16K143 tok/s0.6 sPDF: LightOnOCR 2 1B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit16K143 tok/s0.6 sPDF: LightOnOCR 2 1B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit16K206 tok/s0.3 sPDF: LightOnOCR 2 1B on M4 Max 128 GB report
M5 Max 128 GB8-bit16K219 tok/s0.1 sPDF: LightOnOCR 2 1B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit16K235 tok/s0.2 sPDF: LightOnOCR 2 1B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit16K235 tok/s0.2 sPDF: LightOnOCR 2 1B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit16K235 tok/s0.2 sPDF: LightOnOCR 2 1B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit16K277 tok/s0.1 sPDF: LightOnOCR 2 1B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit16K120 tok/s0.2 sPDF: LightOnOCR 2 1B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit16K120 tok/s0.2 sPDF: LightOnOCR 2 1B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit16K169 tok/s0.1 sPDF: LightOnOCR 2 1B on RTX Spark report
DGX Spark8-bit16K155 tok/s0.1 sPDF: LightOnOCR 2 1B on DGX Spark report
Dual DGX Spark8-bit16K155 tok/s0.1 sPDF: LightOnOCR 2 1B on Dual DGX Spark report
DGX Station GB3008-bit16K948 tok/s0.1 sPDF: LightOnOCR 2 1B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit16K591 tok/s0.1 sPDF: LightOnOCR 2 1B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit16K591 tok/s0.1 sPDF: LightOnOCR 2 1B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit16K591 tok/s0.1 sPDF: LightOnOCR 2 1B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit16K629 tok/s0.1 sPDF: LightOnOCR 2 1B on A100 80 GB SXM report
H100 SXM8-bit16K772 tok/s0.1 sPDF: LightOnOCR 2 1B on H100 SXM report
4× H200 NVL8-bit16K864 tok/s0.1 sPDF: LightOnOCR 2 1B on 4× H200 NVL report
8× H200 NVL8-bit16K864 tok/s0.1 sPDF: LightOnOCR 2 1B 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 noctrex/LightOnOCR-2-1B-GGUF
vLLM
vllm serve lightonai/LightOnOCR-2-1B

Questions

How much VRAM does LightOnOCR 2 1B need?

LightOnOCR 2 1B needs 581 MB for its weights at 4-bit, 2.2 GB with an 8K context and 3.1 GB with 16K. At 8-bit, 2.6 GB.

What is the minimum VRAM for LightOnOCR 2 1B?

About 2.1 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 LightOnOCR 2 1B?

1B parameters: 1.9 GB of weights at 16-bit, 581 MB at 4-bit.

Can LightOnOCR 2 1B run on an RTX 4090?

Yes. At 8-bit it holds 16K of context and writes about 425 tokens a second.

More models