MinerU2.5 Pro 2605 1.2B — Hardware Requirements & GPU Compatibility
VisionMinerU2.5-Pro-2605-1.2B comes from opendatalab as a 1.2-billion-parameter document parsing tool built on the Qwen2-VL architecture, designed to turn PDFs and page images into structured text instead of handling general chat. This version updates the earlier 2604 release; its card notes corrections to layout detection category errors plus better recognition of charts, flowcharts, and seals. For the 2604 model, the card lists an overall OmniDocBench v1.6 score of 95.69, placing it ahead of dedicated OCR models and much larger general vision-language ones, while describing 2605 as only marginally different on that same benchmark. At this scale, it runs fine on a modest consumer GPU or even just a CPU. Context length stands at 8,192 tokens. Released under the Apache 2.0 license, it allows unrestricted commercial and research use, with publication in May 2026. The model processes both Chinese and English documents.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 8K)Best on 24 GB (Q4_K_M · ~642 tok/s · 8K)Macs from 8 GB (Q4_K_M · ~70 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run MinerU2.5 Pro 2605 1.2B
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/mradermacher/MinerU2.5-Pro-2605-1.2B-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 22.9 GB free · 8K context
Download: MinerU2.5-Pro-2605-1.2B.Q4_K_M.gguf0.40 GBmradermacher
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3060 12GB (10 GB headroom) · Q4_K_M
- Generation speed
- ~229 tok/s
- generation speed
- Cost per 1M output tokens
- $0.076
- per 1M output tokens
How Much VRAM Does MinerU2.5 Pro 2605 1.2B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 0.8 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 8K context · ~799 tok/s | |
| 0.8 GB | — | 3-bit small — 3-bit small quantization | Fits: 8K context · ~789 tok/s | |
| 0.9 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 8K context · ~736 tok/s | |
| 0.9 GB | — | 3-bit large — 3-bit large quantization | Fits: 8K context · ~712 tok/s | |
| 1.0 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 8K context · ~642 tok/s | |
| 1.1 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 8K context · ~570 tok/s | |
| 1.3 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 8K context · ~512 tok/s | |
| 1.5 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 8K context · ~443 tok/s |
Selected Q4_K_M
4-bit medium quantization1.1GBat 8K context
- Weights
- 0.7 GB
- KV cache
- 0.1 GB
- Overhead
- 0.3 GB
Which Hardware Can Run MinerU2.5 Pro 2605 1.2B?
Q4_K_M · 1.0 GBMinerU2.5 Pro 2605 1.2B (Q4_K_M) requires 1.0 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 8K context window can add up to 0.1 GB, bringing total usage to 1.1 GB. 114 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 185 devices with unified memory can run MinerU2.5 Pro 2605 1.2B, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB).
≤ 11 GB
≤ 11 GB tier →fits · 2.9 GB free · up to 8K ctx
68 GPUs · 1 Mac class · 3 device families
RTX 3080~485Simulate output on RTX 3080RTX 2080 Ti~393Simulate output on RTX 2080 TiRTX 3070 Ti~388Simulate output on RTX 3070 TiM3*8 GB~70Simulate output on iMac 24" M3 (8 GB)Orin Nano~65Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~47Simulate output on iPhone 1720 GB
fits · 18.9 GB free · up to 8K ctx
1 GPU
RX 7900 XT~471Simulate output on RX 7900 XT32–40 GB
fits · 30.9 GB free · up to 8K ctx
1 GPU · 3 Mac classes · 9 device families
RTX 5090~1142Simulate output on RTX 5090M5 Max*36 GB~316Simulate output on Mac Studio M5 Max (36 GB)M632 GB~117Simulate output on iMac 24" M6 (32 GB)RTX Spark N1X~192Simulate output on NVIDIA RTX Spark N1X Laptop (32 GB)Orin~131Simulate output on NVIDIA Jetson AGX Orin 32GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| MiniCPM5 1BOpenBMB | — | 1.3 GB | ~662 | same size |
| Llama 3.2 1B InstructMeta | — | 2.3 GB | ~799 | same size |
| TinyLlama 1.1B Chat v1.0TinyLlama | — | 1 GB | ~649 | same size |
| LFM2.5 1.2B ThinkingLiquid AI | — | 1.2 GB | ~612 | same size |
| LFM2 1.2BLiquid AI | — | 1.2 GB | ~612 | same size |
| T5gemma L L Ul2 ITGoogle | — | 4.3 GB | ~240 | same size |
Specifications
- Publisher
- opendatalab
- Parameters
- 1.2B
- Architecture
- Qwen2VLForConditionalGeneration
- Context Length
- 8,192 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2026-05-20
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does MinerU2.5 Pro 2605 1.2B need?
MinerU2.5 Pro 2605 1.2B requires 1.0 GB of VRAM at Q4_K_M, or 2.6 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 1.2B × 4.8 bits ÷ 8 = 0.7 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 ≈ 0.4 GB (at full 8K context)
VRAM usage by quantization
What's the best quantization for MinerU2.5 Pro 2605 1.2B?
For MinerU2.5 Pro 2605 1.2B, Q4_K_M (1.0 GB) offers the best balance of quality and VRAM usage. Q5_K_S (1.1 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 0.8 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run MinerU2.5 Pro 2605 1.2B on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run MinerU2.5 Pro 2605 1.2B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 1.0 GB of usable unified memory (RAM minus macOS overhead).
Can I run MinerU2.5 Pro 2605 1.2B locally?
Yes — MinerU2.5 Pro 2605 1.2B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is MinerU2.5 Pro 2605 1.2B?
At Q4_K_M, MinerU2.5 Pro 2605 1.2B can reach ~4706 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~642 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 ÷ 1.02 × 0.65 = ~5098 tok/s
Estimated speed at Q4_K_M (1.0 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of MinerU2.5 Pro 2605 1.2B?
At Q4_K_M, the download is about 0.69 GB. The full-precision BF16 version is 2.31 GB. The smallest option (Q2_K) is 0.49 GB.
Which GPUs can run MinerU2.5 Pro 2605 1.2B?
114 consumer GPUs can run MinerU2.5 Pro 2605 1.2B at Q4_K_M. Top options include AMD Radeon RX 5600 XT, AMD Radeon RX 5700 XT, AMD Radeon RX 580 8GB. 114 GPUs have plenty of headroom for comfortable inference.
Which devices can run MinerU2.5 Pro 2605 1.2B?
185 devices with unified memory can run MinerU2.5 Pro 2605 1.2B at Q4_K_M (1.0 GB), including AMD Ryzen 7 8845HS (Hawk Point) Laptop, AMD Ryzen AI 7 350 (Krackan Point) Laptop, AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.