Jina Ocr V1 — Hardware Requirements & GPU Compatibility
Vision
jina-ocr-v1 is Jina AI's end-to-end document-parsing OCR model, built on the DeepSeek-OCR architecture. It combines a DeepEncoder vision tower, which represents a 1024x1024 page as just 256 visual tokens plus dynamic local tiles, with a 3-billion-parameter Mixture-of-Experts decoder, and adds a FastMTP speculative-decoding head that drafts several tokens at once to speed up long structured outputs. It reads documents into clean Markdown, converts equations to LaTeX and tables to HTML, and scores ahead of DeepSeek-OCR on the OmniDocBench and olmOCR-Bench document-parsing benchmarks. Its small active-parameter footprint lets it run on a single modest consumer GPU. Context length is 32,768 tokens. It is released under the CC BY-NC 4.0 license, which restricts use to non-commercial purposes. It was published in September 2026.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 32K)Best on 24 GB (Q4_K_M · ~509 tok/s · 32K)Macs from 8 GB (Q4_K_M · ~70 tok/s)
CC BY-NC 4.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Jina Ocr V1
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/ahmadw/jina-ocr-v1-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 20.7 GB free · 16K context
Download: jina-ocr-v1-Q4_K_M.gguf1.95 GBahmadw
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 24 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on RTX 3060 12GB (8 GB headroom) · Q4_K_M
- Generation speed
- ~232 tok/s
- generation speed
- Cost per 1M output tokens
- $0.074
- per 1M output tokens
How Much VRAM Does Jina Ocr V1 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 1.9 GB | 3.8 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 32K context · ~606 tok/s | |
| 2.1 GB | 4 GB | 3-bit medium — 3-bit medium quantization | Fits: 32K context · ~567 tok/s | |
| 2.5 GB | 4.3 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 32K context · ~509 tok/s | |
| 2.8 GB | 4.7 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 32K context · ~461 tok/s | |
| 3.2 GB | 5.1 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 32K context · ~422 tok/s | |
| 3.8 GB | 5.7 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 32K context · ~373 tok/s | |
| 7.2 GB | 9.1 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 32K context · ~223 tok/s |
Selected Q4_K_M
4-bit medium quantization3.3GBat 16K context
- Weights
- 2.0 GB
- KV cache
- 1.0 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 20.7 GB free · ~509 tok/s
≥50 tok/s needs ≥68 GB/s · RTX 5090 ~717, RTX 3090 Ti ~509
Which Hardware Can Run Jina Ocr V1?
Q4_K_M · 2.5 GBJina Ocr V1 (Q4_K_M) requires 2.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 4+ GB is recommended. Using the full 33K context window can add up to 1.9 GB, bringing total usage to 4.3 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Jina Ocr V1, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 4.7 GB free · up to 32K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~418Simulate output on RTX 3080RTX 2080 Ti~358Simulate output on RTX 2080 TiRTX 3070 Ti~354Simulate output on RTX 3070 TiM3*8 GB~70Simulate output on iMac 24" M3 (8 GB)Orin Nano~74Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~49Simulate output on iPhone 1720 GB
fits · 16.7 GB free · up to 32K ctx
1 GPU
RX 7900 XT~239Simulate output on RX 7900 XT32–40 GB
fits · 28.7 GB free · up to 32K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~717Simulate output on RTX 5090M5 Max36 GB~198Simulate output on Mac Studio M5 Max (36 GB)M532 GB~97Simulate output on MacBook Air 13" M5 (32 GB)Orin~141Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~91Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| DeepSeek OCR 2DeepSeek | — | 2.8 GB | ~504 | same size |
| Granite 4.0 MicroIBM | — | 3.7 GB | ~261 | same size |
| Llama 3.2 3B InstructMeta | — | 4.1 GB | ~309 | same size |
| Tiny Aya GlobalCohere | — | 9.4 GB | ~89 | same size |
| AI21 Jamba Reasoning 3BAI21 Labs | — | 2.3 GB | ~291 | same size |
| MiniMax M3 EAGLE3Inferact | — | 2.7 GB | ~284 | same size |
Specifications
- Publisher
- jinaai
- Parameters
- 3.4B
- Active parameters
- 1.2B per token
- Architecture
- DeepseekOCRForCausalLM
- Context Length
- 32,768 tokens
- Vocabulary Size
- 129,280
- Release Date
- 2026-09-01
- License
- CC BY-NC 4.0
Frequently Asked Questions
How much VRAM does Jina Ocr V1 need?
Jina Ocr V1 requires 2.5 GB of VRAM at Q4_K_M, or 7.2 GB at BF16. Full 33K context adds up to 1.9 GB (4.3 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 3.4B × 4.8 bits ÷ 8 = 2 GB
KV Cache + Overhead ≈ 0.5 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 ≈ 2.3 GB (at full 33K context)
VRAM usage by quantization
What's the best quantization for Jina Ocr V1?
For Jina Ocr V1, Q4_K_M (2.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (2.8 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 1.9 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Jina Ocr V1 on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Jina Ocr V1. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 2.5 GB of usable unified memory (RAM minus macOS overhead).
Can I run Jina Ocr V1 locally?
Yes — Jina Ocr V1 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Jina Ocr V1?
At Q4_K_M, Jina Ocr V1 can reach ~388 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~509 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 0.9 GB active ÷ (8000 × 0.65) = 0.1645 ms, plus 12 layers × 0.055 ms = 0.6600 ms, so 1000 ÷ 0.8245 ms = ~1213 tok/s
Estimated speed at Q4_K_M (2.5 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Jina Ocr V1?
At Q4_K_M, the download is about 2.02 GB. The full-precision BF16 version is 6.74 GB. The smallest option (Q2_K) is 1.43 GB.
Which GPUs can run Jina Ocr V1?
52 consumer GPUs can run Jina Ocr V1 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 Jina Ocr V1?
163 devices with unified memory can run Jina Ocr V1 at Q4_K_M (2.5 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.