Skip to content

TeleOCR VRAM requirements

How much VRAM TeleOCR needs
4-bit weights: 818 MBWith 8K context: 2.4 GBWith 32K context: 5.1 GB8-bit, 8K context: 3.0 GB
Parameters
1.4B
Longest context
125K
Released
Aug 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM TeleOCR needs

Memory TeleOCR needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 125K
16-bit2.6 GB4.3 GB6.9 GB17.1 GB
8-bit1.4 GB3.0 GB5.7 GB15.8 GB
6-bit1.1 GB2.7 GB5.4 GB15.5 GB
5-bit960 MB2.6 GB5.2 GB15.4 GB
4-bit818 MB2.4 GB5.1 GB15.2 GB
3-bit660 MB2.3 GB4.9 GB15.1 GB
2-bit506 MB2.1 GB4.8 GB14.9 GB
1.6-bit278 MB1.9 GB4.5 GB14.7 GB

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

GPUs and Macs that run TeleOCR

The best way TeleOCR runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit48K114 tok/s0.2 sPDF: TeleOCR on RTX 4060 report
RTX 50608-bit48K177 tok/s0.1 sPDF: TeleOCR on RTX 5060 report
RTX 30708-bit48K177 tok/s0.1 sPDF: TeleOCR on RTX 3070 report
RTX 3080 10 GB8-bit64K273 tok/s0.1 sPDF: TeleOCR on RTX 3080 10 GB report
RTX 3060 12 GB8-bit80K147 tok/s0.2 sPDF: TeleOCR on RTX 3060 12 GB report
Arc B5808-bit80K121 tok/s0.2 sPDF: TeleOCR on Arc B580 report
RTX 40708-bit80K196 tok/s0.1 sPDF: TeleOCR on RTX 4070 report
RTX 4070 Super8-bit80K196 tok/s0.1 sPDF: TeleOCR on RTX 4070 Super report
RTX 50708-bit80K248 tok/s0.1 sPDF: TeleOCR on RTX 5070 report
RTX 4060 Ti 16 GB8-bit112K121 tok/s0.1 sPDF: TeleOCR on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit112K177 tok/s0.1 sPDF: TeleOCR on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit112K142 tok/s0.1 sPDF: TeleOCR on Arc A770 16 GB report
Radeon RX 9070 XT8-bit112K188 tok/s0.1 sPDF: TeleOCR on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit112K248 tok/s0.1 sPDF: TeleOCR on RTX 4070 Ti Super report
RTX 4080 Super8-bit112K266 tok/s0.1 sPDF: TeleOCR on RTX 4080 Super report
RTX 5070 Ti8-bit112K309 tok/s0.1 sPDF: TeleOCR on RTX 5070 Ti report
RTX 50808-bit112K325 tok/s0.1 sPDF: TeleOCR on RTX 5080 report
Radeon RX 7900 XTX8-bit125K249 tok/s0.1 sPDF: TeleOCR on Radeon RX 7900 XTX report
RTX 30908-bit125K319 tok/s0.1 sPDF: TeleOCR on RTX 3090 report
RTX 40908-bit125K337 tok/s0.1 sPDF: TeleOCR on RTX 4090 report
Radeon AI PRO R97008-bit125K188 tok/s0.1 sPDF: TeleOCR on Radeon AI PRO R9700 report
RTX 50908-bit125K491 tok/s0.1 sPDF: TeleOCR on RTX 5090 report
2× RTX 30908-bit125K319 tok/s0.1 sPDF: TeleOCR on 2× RTX 3090 report
2× RTX 40908-bit125K337 tok/s0.1 sPDF: TeleOCR on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit125K410 tok/s0.1 sPDF: TeleOCR on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit125K325 tok/s0.1 sPDF: TeleOCR on RTX 6000 Ada report
L40S8-bit125K301 tok/s0.1 sPDF: TeleOCR on L40S report
2× RTX 50908-bit125K491 tok/s0.1 sPDF: TeleOCR on 2× RTX 5090 report
4× RTX 30908-bit125K319 tok/s0.1 sPDF: TeleOCR on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit125K491 tok/s0.1 sPDF: TeleOCR on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit72K57 tok/s1.8 sPDF: TeleOCR on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit120K57 tok/s1.8 sPDF: TeleOCR on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit125K57 tok/s1.8 sPDF: TeleOCR on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit120K112 tok/s0.9 sPDF: TeleOCR on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit125K112 tok/s0.9 sPDF: TeleOCR on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit125K112 tok/s0.9 sPDF: TeleOCR on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit125K169 tok/s0.5 sPDF: TeleOCR on M4 Max 128 GB report
M5 Max 128 GB8-bit125K182 tok/s0.1 sPDF: TeleOCR on M5 Max 128 GB report
M3 Ultra 96 GB8-bit125K197 tok/s0.3 sPDF: TeleOCR on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit125K197 tok/s0.3 sPDF: TeleOCR on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit125K197 tok/s0.3 sPDF: TeleOCR on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit125K240 tok/s0.1 sPDF: TeleOCR on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit125K89 tok/s0.3 sPDF: TeleOCR on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit125K89 tok/s0.3 sPDF: TeleOCR on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit125K125 tok/s0.1 sPDF: TeleOCR on RTX Spark report
DGX Spark8-bit125K115 tok/s0.1 sPDF: TeleOCR on DGX Spark report
Dual DGX Spark8-bit125K115 tok/s0.1 sPDF: TeleOCR on Dual DGX Spark report
DGX Station GB3008-bit125K875 tok/s0.1 sPDF: TeleOCR on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit125K491 tok/s0.1 sPDF: TeleOCR on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit125K491 tok/s0.1 sPDF: TeleOCR on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit125K491 tok/s0.1 sPDF: TeleOCR on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit125K528 tok/s0.1 sPDF: TeleOCR on A100 80 GB SXM report
H100 SXM8-bit125K675 tok/s0.1 sPDF: TeleOCR on H100 SXM report
4× H200 NVL8-bit125K777 tok/s0.1 sPDF: TeleOCR on 4× H200 NVL report
8× H200 NVL8-bit125K777 tok/s0.1 sPDF: TeleOCR 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 cstr/TeleOCR-GGUF
vLLM
vllm serve XingChen-AGI/TeleOCR

Questions

How much VRAM does TeleOCR need?

TeleOCR needs 818 MB for its weights at 4-bit, 2.4 GB with an 8K context and 5.1 GB with 32K. At 8-bit, 3.0 GB.

What is the minimum VRAM for TeleOCR?

About 2.3 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 RTX 4060.

How big is TeleOCR?

1.4B parameters: 2.6 GB of weights at 16-bit, 818 MB at 4-bit.

Can TeleOCR run on an RTX 4090?

Yes. At 8-bit it holds 125K of context and writes about 337 tokens a second.

More models