PaddleOCR VL 1.6 — Hardware Requirements & GPU Compatibility
Vision
PaddleOCR-VL-1.6 is PaddlePaddle's compact, roughly 0.9-billion-parameter vision-language model for document parsing, built on the ERNIE 4.5 line and specialized for OCR, table, formula, chart, and seal/stamp recognition plus text spotting rather than open-domain chat. It upgrades PaddleOCR-VL-1.5 with a region-aware data optimization framework that targets the earlier model's weak spots and a progressive post-training recipe combining curated data selection with reinforcement learning, while staying architecture-compatible with 1.5 for drop-in migration. The card reports a new state-of-the-art 96.33% on OmniDocBench v1.6. At under a billion parameters it runs on a single modest consumer GPU. Context length is 131,072 tokens. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in May 2026.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 128K)Best on 24 GB (Q4_K_M · ~736 tok/s · 128K)Macs from 8 GB (Q4_K_M · ~81 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run PaddleOCR VL 1.6
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
ollama run hf.co/LunarOilRig/PaddleOCR-VL-1.6-GGUF-Q4:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 23.0 GB free · 16K context
Download: PaddleOCR-VL-1.6-Q4_K_M.gguf0.30 GBLunarOilRig
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (11 GB headroom) · Q4_K_M
- Generation speed
- ~666 tok/s
- generation speed
- Cost per 1M output tokens
- $0.051
- per 1M output tokens
How Much VRAM Does PaddleOCR VL 1.6 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 0.7 GB | 1 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 128K context · ~898 tok/s | |
| 0.7 GB | 1 GB | 3-bit small — 3-bit small quantization | Fits: 128K context · ~885 tok/s | |
| 0.8 GB | 1.1 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~829 tok/s | |
| 0.8 GB | 1.1 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 128K context · ~819 tok/s | |
| 0.9 GB | 1.2 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 128K context · ~736 tok/s | |
| 1 GB | 1.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 128K context · ~655 tok/s | |
| 1.1 GB | 1.4 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 128K context · ~590 tok/s | |
| 1.3 GB | 1.6 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 128K context · ~512 tok/s |
Selected Q4_K_M
4-bit medium quantization1.0GBat 16K context
- Weights
- 0.6 GB
- KV cache
- 0.2 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 23.0 GB free · ~736 tok/s
≥50 tok/s needs ≥68 GB/s · RTX 5090 ~1309, RTX 3090 Ti ~736
Which Hardware Can Run PaddleOCR VL 1.6?
Q4_K_M · 0.9 GBPaddleOCR VL 1.6 (Q4_K_M) requires 0.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 2+ GB is recommended. Using the full 131K context window can add up to 1.2 GB, bringing total usage to 2.1 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run PaddleOCR VL 1.6, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 7.0 GB free · up to 128K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~555Simulate output on RTX 3080RTX 2080 Ti~450Simulate output on RTX 2080 TiRTX 3070 Ti~444Simulate output on RTX 3070 TiM3*8 GB~81Simulate output on iMac 24" M3 (8 GB)Orin Nano~75Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~54Simulate output on iPhone 1720 GB
fits · 19.0 GB free · up to 128K ctx
1 GPU
RX 7900 XT~539Simulate output on RX 7900 XT32–40 GB
fits · 31.0 GB free · up to 128K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~1309Simulate output on RTX 5090M5 Max36 GB~362Simulate output on Mac Studio M5 Max (36 GB)M532 GB~121Simulate output on MacBook Air 13" M5 (32 GB)Orin~150Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~99Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
3| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| NVIDIA Nemotron Parse 2.0NVIDIA | — | 3.4 GB | ~329 | same size |
| Gemma 3 1B ITGoogle | — | 2.1 GB | ~993 | same size |
| Qwen3 8B Speculator.eagle3RedHatAI | — | 2.1 GB | ~978 | same size |
Specifications
- Publisher
- PaddlePaddle
- Parameters
- 959M
- Architecture
- PaddleOCRVLForConditionalGeneration
- Context Length
- 131,072 tokens
- Vocabulary Size
- 103,424
- Release Date
- 2026-05-27
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does PaddleOCR VL 1.6 need?
PaddleOCR VL 1.6 requires 0.9 GB of VRAM at Q4_K_M, or 2.2 GB at BF16. Full 131K context adds up to 1.2 GB (2.1 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 959M × 4.8 bits ÷ 8 = 0.6 GB
KV Cache + Overhead ≈ 0.3 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 1.5 GB (at full 131K context)
VRAM usage by quantization
What's the best quantization for PaddleOCR VL 1.6?
For PaddleOCR VL 1.6, Q4_K_M (0.9 GB) offers the best balance of quality and VRAM usage. Q5_K_S (1.0 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 0.6 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run PaddleOCR VL 1.6 on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run PaddleOCR VL 1.6. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.9 GB of usable unified memory (RAM minus macOS overhead).
Can I run PaddleOCR VL 1.6 locally?
Yes — PaddleOCR VL 1.6 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is PaddleOCR VL 1.6?
At Q4_K_M, PaddleOCR VL 1.6 can reach ~5393 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~736 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 0.89 × 0.65 = ~5843 tok/s
Estimated speed at Q4_K_M (0.9 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of PaddleOCR VL 1.6?
At Q4_K_M, the download is about 0.58 GB. The full-precision BF16 version is 1.92 GB. The smallest option (IQ2_XXS) is 0.26 GB.
Which GPUs can run PaddleOCR VL 1.6?
52 consumer GPUs can run PaddleOCR VL 1.6 at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run PaddleOCR VL 1.6?
163 devices with unified memory can run PaddleOCR VL 1.6 at Q4_K_M (0.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.