All tasks / Text & knowledge / Documents & OCR
Best local model for documents & ocr on a 8 GB GPU
Open-weights models whose approximate minimum GPU memory is 8 GB or less, ranked from public benchmarks. Memory figures are approximate; quantised builds can need less.
STop ranked: TeleOCR (XingChen-AGI), 97 on OmniDocBench v1.6 (end-to-end, v1.6_full).8 options listed · data verified 19 Oct 2026
Ranked list
| Tier | Name | Approach | Runs | Rated by | Price | Skill | Compare |
|---|---|---|---|---|---|---|---|
| S |  TeleOCRXingChen-AGI | Models | Local | 97 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~5 GB VRAM | — | |
| S |  OvisOCR2ATH-MaaS | Models | Local | 96 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~4 GB VRAM | — | vs TeleOCR |
| S |  PaddleOCR-VL-1.6Baidu (PaddlePaddle) | Models | Local | 96 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~4 GB VRAM | — | vs TeleOCR |
| S |  MinerU2.5-Pro-2605-1.2BOpenDataLab (Shanghai AI Laboratory) | Models | Local | 96 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~4 GB VRAM | — | vs TeleOCR |
| S |  GLM-OCRZ.ai | Models | Local | 95 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~5 GB VRAM | — | vs TeleOCR |
| S |  Unlimited-OCRBaidu | Models | Local | 94 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~8 GB VRAM | — | |
| S |  Youtu-ParsingTencent (Youtu) | Models | Local | 94 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~7 GB VRAM | — | |
| S |  FireRed-OCRFireRed Team | Models | Local | 93 on OmniDocBench v1.6 (end-to-end, v1.6_full) | ~6 GB VRAM | — |
Tiers come only from public benchmarks (OmniDocBench v1.6 (end-to-end, v1.6_full)); tools without a result of their own inherit the tier of the best model they offer. How we rank.