TeleOCR VRAM requirements
| 4-bit weights: 818 MB | With 8K context: 2.4 GB | With 32K context: 5.1 GB | 8-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
| Precision | Weights | With 8K | With 32K | With 125K |
|---|---|---|---|---|
| 16-bit | 2.6 GB | 4.3 GB | 6.9 GB | 17.1 GB |
| 8-bit | 1.4 GB | 3.0 GB | 5.7 GB | 15.8 GB |
| 6-bit | 1.1 GB | 2.7 GB | 5.4 GB | 15.5 GB |
| 5-bit | 960 MB | 2.6 GB | 5.2 GB | 15.4 GB |
| 4-bit | 818 MB | 2.4 GB | 5.1 GB | 15.2 GB |
| 3-bit | 660 MB | 2.3 GB | 4.9 GB | 15.1 GB |
| 2-bit | 506 MB | 2.1 GB | 4.8 GB | 14.9 GB |
| 1.6-bit | 278 MB | 1.9 GB | 4.5 GB | 14.7 GB |
One conversation, 16-bit context cache, runtime overhead included.
GPUs and Macs that run TeleOCR
| Machine | Best fit | Context | Speed | First reply | Report |
|---|---|---|---|---|---|
| RTX 4060 | 8-bit | 48K | 114 tok/s | 0.2 s | PDF: TeleOCR on RTX 4060 report |
| RTX 5060 | 8-bit | 48K | 177 tok/s | 0.1 s | PDF: TeleOCR on RTX 5060 report |
| RTX 3070 | 8-bit | 48K | 177 tok/s | 0.1 s | PDF: TeleOCR on RTX 3070 report |
| RTX 3080 10 GB | 8-bit | 64K | 273 tok/s | 0.1 s | PDF: TeleOCR on RTX 3080 10 GB report |
| RTX 3060 12 GB | 8-bit | 80K | 147 tok/s | 0.2 s | PDF: TeleOCR on RTX 3060 12 GB report |
| Arc B580 | 8-bit | 80K | 121 tok/s | 0.2 s | PDF: TeleOCR on Arc B580 report |
| RTX 4070 | 8-bit | 80K | 196 tok/s | 0.1 s | PDF: TeleOCR on RTX 4070 report |
| RTX 4070 Super | 8-bit | 80K | 196 tok/s | 0.1 s | PDF: TeleOCR on RTX 4070 Super report |
| RTX 5070 | 8-bit | 80K | 248 tok/s | 0.1 s | PDF: TeleOCR on RTX 5070 report |
| RTX 4060 Ti 16 GB | 8-bit | 112K | 121 tok/s | 0.1 s | PDF: TeleOCR on RTX 4060 Ti 16 GB report |
| RTX 5060 Ti 16 GB | 8-bit | 112K | 177 tok/s | 0.1 s | PDF: TeleOCR on RTX 5060 Ti 16 GB report |
| Arc A770 16 GB | 8-bit | 112K | 142 tok/s | 0.1 s | PDF: TeleOCR on Arc A770 16 GB report |
| Radeon RX 9070 XT | 8-bit | 112K | 188 tok/s | 0.1 s | PDF: TeleOCR on Radeon RX 9070 XT report |
| RTX 4070 Ti Super | 8-bit | 112K | 248 tok/s | 0.1 s | PDF: TeleOCR on RTX 4070 Ti Super report |
| RTX 4080 Super | 8-bit | 112K | 266 tok/s | 0.1 s | PDF: TeleOCR on RTX 4080 Super report |
| RTX 5070 Ti | 8-bit | 112K | 309 tok/s | 0.1 s | PDF: TeleOCR on RTX 5070 Ti report |
| RTX 5080 | 8-bit | 112K | 325 tok/s | 0.1 s | PDF: TeleOCR on RTX 5080 report |
| Radeon RX 7900 XTX | 8-bit | 125K | 249 tok/s | 0.1 s | PDF: TeleOCR on Radeon RX 7900 XTX report |
| RTX 3090 | 8-bit | 125K | 319 tok/s | 0.1 s | PDF: TeleOCR on RTX 3090 report |
| RTX 4090 | 8-bit | 125K | 337 tok/s | 0.1 s | PDF: TeleOCR on RTX 4090 report |
| Radeon AI PRO R9700 | 8-bit | 125K | 188 tok/s | 0.1 s | PDF: TeleOCR on Radeon AI PRO R9700 report |
| RTX 5090 | 8-bit | 125K | 491 tok/s | 0.1 s | PDF: TeleOCR on RTX 5090 report |
| 2× RTX 3090 | 8-bit | 125K | 319 tok/s | 0.1 s | PDF: TeleOCR on 2× RTX 3090 report |
| 2× RTX 4090 | 8-bit | 125K | 337 tok/s | 0.1 s | PDF: TeleOCR on 2× RTX 4090 report |
| RTX PRO 5000 Blackwell 48 GB | 8-bit | 125K | 410 tok/s | 0.1 s | PDF: TeleOCR on RTX PRO 5000 Blackwell 48 GB report |
| RTX 6000 Ada | 8-bit | 125K | 325 tok/s | 0.1 s | PDF: TeleOCR on RTX 6000 Ada report |
| L40S | 8-bit | 125K | 301 tok/s | 0.1 s | PDF: TeleOCR on L40S report |
| 2× RTX 5090 | 8-bit | 125K | 491 tok/s | 0.1 s | PDF: TeleOCR on 2× RTX 5090 report |
| 4× RTX 3090 | 8-bit | 125K | 319 tok/s | 0.1 s | PDF: TeleOCR on 4× RTX 3090 report |
| RTX PRO 6000 Blackwell | 8-bit | 125K | 491 tok/s | 0.1 s | PDF: TeleOCR on RTX PRO 6000 Blackwell report |
| Mac mini M4 16 GB | 8-bit | 72K | 57 tok/s | 1.8 s | PDF: TeleOCR on Mac mini M4 16 GB report |
| Mac mini M4 24 GB | 8-bit | 120K | 57 tok/s | 1.8 s | PDF: TeleOCR on Mac mini M4 24 GB report |
| Mac mini M4 32 GB | 8-bit | 125K | 57 tok/s | 1.8 s | PDF: TeleOCR on Mac mini M4 32 GB report |
| Mac mini M4 Pro 24 GB | 8-bit | 120K | 112 tok/s | 0.9 s | PDF: TeleOCR on Mac mini M4 Pro 24 GB report |
| Mac mini M4 Pro 48 GB | 8-bit | 125K | 112 tok/s | 0.9 s | PDF: TeleOCR on Mac mini M4 Pro 48 GB report |
| Mac mini M4 Pro 64 GB | 8-bit | 125K | 112 tok/s | 0.9 s | PDF: TeleOCR on Mac mini M4 Pro 64 GB report |
| M4 Max 128 GB | 8-bit | 125K | 169 tok/s | 0.5 s | PDF: TeleOCR on M4 Max 128 GB report |
| M5 Max 128 GB | 8-bit | 125K | 182 tok/s | 0.1 s | PDF: TeleOCR on M5 Max 128 GB report |
| M3 Ultra 96 GB | 8-bit | 125K | 197 tok/s | 0.3 s | PDF: TeleOCR on M3 Ultra 96 GB report |
| M3 Ultra 256 GB | 8-bit | 125K | 197 tok/s | 0.3 s | PDF: TeleOCR on M3 Ultra 256 GB report |
| M3 Ultra 512 GB | 8-bit | 125K | 197 tok/s | 0.3 s | PDF: TeleOCR on M3 Ultra 512 GB report |
| M5 Ultra 512 GB | 8-bit | 125K | 240 tok/s | 0.1 s | PDF: TeleOCR on M5 Ultra 512 GB report |
| Ryzen AI Max+ 395 64 GB | 8-bit | 125K | 89 tok/s | 0.3 s | PDF: TeleOCR on Ryzen AI Max+ 395 64 GB report |
| Ryzen AI Max+ 395 128 GB | 8-bit | 125K | 89 tok/s | 0.3 s | PDF: TeleOCR on Ryzen AI Max+ 395 128 GB report |
| RTX Spark | 8-bit | 125K | 125 tok/s | 0.1 s | PDF: TeleOCR on RTX Spark report |
| DGX Spark | 8-bit | 125K | 115 tok/s | 0.1 s | PDF: TeleOCR on DGX Spark report |
| Dual DGX Spark | 8-bit | 125K | 115 tok/s | 0.1 s | PDF: TeleOCR on Dual DGX Spark report |
| DGX Station GB300 | 8-bit | 125K | 875 tok/s | 0.1 s | PDF: TeleOCR on DGX Station GB300 report |
| 2× RTX PRO 6000 Blackwell | 8-bit | 125K | 491 tok/s | 0.1 s | PDF: TeleOCR on 2× RTX PRO 6000 Blackwell report |
| 4× RTX PRO 6000 Blackwell | 8-bit | 125K | 491 tok/s | 0.1 s | PDF: TeleOCR on 4× RTX PRO 6000 Blackwell report |
| 8× RTX PRO 6000 Blackwell | 8-bit | 125K | 491 tok/s | 0.1 s | PDF: TeleOCR on 8× RTX PRO 6000 Blackwell report |
| A100 80 GB SXM | 8-bit | 125K | 528 tok/s | 0.1 s | PDF: TeleOCR on A100 80 GB SXM report |
| H100 SXM | 8-bit | 125K | 675 tok/s | 0.1 s | PDF: TeleOCR on H100 SXM report |
| 4× H200 NVL | 8-bit | 125K | 777 tok/s | 0.1 s | PDF: TeleOCR on 4× H200 NVL report |
| 8× H200 NVL | 8-bit | 125K | 777 tok/s | 0.1 s | PDF: 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.