Skip to content

LightOnOCR 3 0.8B VRAM requirements

How much VRAM LightOnOCR 3 0.8B needs
4-bit weights: 493 MBWith 8K context: 1.5 GBWith 32K context: 1.8 GB8-bit, 8K context: 1.9 GB
Parameters
853M
Longest context
256K
Released
Oct 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM LightOnOCR 3 0.8B needs

Memory LightOnOCR 3 0.8B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit1.6 GB2.7 GB2.9 GB4.1 GB
8-bit864 MB1.9 GB2.2 GB3.3 GB
6-bit667 MB1.7 GB2.0 GB3.1 GB
5-bit579 MB1.6 GB1.9 GB3.0 GB
4-bit493 MB1.5 GB1.8 GB3.0 GB
3-bit398 MB1.5 GB1.7 GB2.9 GB
2-bit305 MB1.4 GB1.6 GB2.8 GB
1.6-bit168 MB1.2 GB1.5 GB2.6 GB

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

GPUs and Macs that run LightOnOCR 3 0.8B

The best way LightOnOCR 3 0.8B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit256K179 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4060 report
RTX 50608-bit256K272 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 5060 report
RTX 30708-bit256K272 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 3070 report
RTX 3080 10 GB8-bit256K406 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit256K228 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 3060 12 GB report
Arc B5808-bit256K182 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Arc B580 report
RTX 40708-bit256K299 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4070 report
RTX 4070 Super8-bit256K299 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4070 Super report
RTX 50708-bit256K372 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit256K188 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit256K272 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit256K211 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit256K277 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit256K372 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4070 Ti Super report
RTX 4080 Super8-bit256K397 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4080 Super report
RTX 5070 Ti8-bit256K455 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 5070 Ti report
RTX 50808-bit256K477 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 5080 report
Radeon RX 7900 XTX8-bit256K356 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Radeon RX 7900 XTX report
RTX 30908-bit256K469 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 3090 report
RTX 40908-bit256K492 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 4090 report
Radeon AI PRO R97008-bit256K277 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Radeon AI PRO R9700 report
RTX 50908-bit256K686 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 5090 report
2× RTX 30908-bit256K469 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 2× RTX 3090 report
2× RTX 40908-bit256K492 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K587 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K477 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX 6000 Ada report
L40S8-bit256K444 tok/s0.1 sPDF: LightOnOCR 3 0.8B on L40S report
2× RTX 50908-bit256K686 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 2× RTX 5090 report
4× RTX 30908-bit256K469 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K686 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit256K88 tok/s0.9 sPDF: LightOnOCR 3 0.8B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit256K88 tok/s0.9 sPDF: LightOnOCR 3 0.8B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit256K88 tok/s0.9 sPDF: LightOnOCR 3 0.8B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit256K165 tok/s0.4 sPDF: LightOnOCR 3 0.8B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit256K165 tok/s0.4 sPDF: LightOnOCR 3 0.8B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K165 tok/s0.4 sPDF: LightOnOCR 3 0.8B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K239 tok/s0.2 sPDF: LightOnOCR 3 0.8B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K254 tok/s0.1 sPDF: LightOnOCR 3 0.8B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K273 tok/s0.1 sPDF: LightOnOCR 3 0.8B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K273 tok/s0.1 sPDF: LightOnOCR 3 0.8B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K273 tok/s0.1 sPDF: LightOnOCR 3 0.8B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K322 tok/s0.1 sPDF: LightOnOCR 3 0.8B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K138 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K138 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K195 tok/s0.1 sPDF: LightOnOCR 3 0.8B on RTX Spark report
DGX Spark8-bit256K180 tok/s0.1 sPDF: LightOnOCR 3 0.8B on DGX Spark report
Dual DGX Spark8-bit256K180 tok/s0.1 sPDF: LightOnOCR 3 0.8B on Dual DGX Spark report
DGX Station GB3008-bit256K1,103 tok/s0.1 sPDF: LightOnOCR 3 0.8B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K686 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K686 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K686 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K731 tok/s0.1 sPDF: LightOnOCR 3 0.8B on A100 80 GB SXM report
H100 SXM8-bit256K897 tok/s0.1 sPDF: LightOnOCR 3 0.8B on H100 SXM report
4× H200 NVL8-bit256K1,004 tok/s0.1 sPDF: LightOnOCR 3 0.8B on 4× H200 NVL report
8× H200 NVL8-bit256K1,004 tok/s0.1 sPDF: LightOnOCR 3 0.8B 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 prithivMLmods/LightOnOCR-3-0.8B-GGUF
vLLM
vllm serve lightonai/LightOnOCR-3-0.8B

Questions

How much VRAM does LightOnOCR 3 0.8B need?

LightOnOCR 3 0.8B needs 493 MB for its weights at 4-bit, 1.5 GB with an 8K context and 1.8 GB with 32K. At 8-bit, 1.9 GB.

What is the minimum VRAM for LightOnOCR 3 0.8B?

About 1.5 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 4060.

How big is LightOnOCR 3 0.8B?

853M parameters: 1.6 GB of weights at 16-bit, 493 MB at 4-bit.

Can LightOnOCR 3 0.8B run on an RTX 4090?

Yes. At 8-bit it holds 256K of context and writes about 492 tokens a second.

More models