InternVL3.5 vs Phi-4-multimodal
Pick InternVL3.5 for self-hosted document question answering, especially when you need non-English documents or a choice of model sizes from 1B to 241B-A28B. Pick Phi-4-multimodal for English document Q&A on a single modest GPU, or when audio and images go through one model. The main split is language: Phi-4-multimodal supports vision for English only, while InternVL3.5 has no such stated limit.
InternVL3.5opengvlab/internvl | Phi-4-multimodalmicrosoft/phi-4-multimodal | |
|---|---|---|
| Type | Vision model | Vision model |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 3.0 s / page | 2.5 s / page |
| Best for | Self-hosted document understanding | Small model for OCR and questions |
| Inputs | images | images |
| Languages | Multilingual | English for vision |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose InternVL3.5 if
- It comes in nine sizes from 1B to 241B, so you can trade accuracy for cost.
- Phi-4-multimodal's model card says vision is English only and that quality drops in other languages, while InternVL3.5's card states no such limit.
- The MoE variants 20B-A4B and 30B-A3B activate few parameters per token, and models up to 30B fit on a single A100.
Choose Phi-4-multimodal if
- It scores 93.2 on DocVQA and 84.4 on OCRBench at 5.6B parameters, as reported by Microsoft.
- It takes audio as well as images, so one model can serve mixed audio and image pipelines.
- It has a 128K token context and is also available through Azure AI Foundry, in addition to self-hosting.
Key differences
| InternVL3.5 | Phi-4-multimodal | |
|---|---|---|
| Licence | Apache-2.0 | MIT |
| How you run it | Self-host with Transformers, LMDeploy or vLLM | Self-host or use Azure AI Foundry |
| Model size | 1B to 241B-A28B, nine sizes | Single 5.6B model |
| Hardware | Up to 30B on one A100; 38B needs two A100s; 241B-A28B needs eight A100s | Flash attention tested on A100, A6000 and H100; V100 and older need eager attention |
| Published OCR scores | Only chart images on the model card, no text values | DocVQA 93.2, OCRBench 84.4, ChartQA 81.4, InfoVQA 72.7 (authors' numbers) |
| Input types | Image and text | Text, image and audio |
| Known limits | No bounding boxes or layout output by default; the 241B-A28B model needs eight A100s | English-only vision; may give repetitive or inconsistent output in long chats; no coordinates or layout |
Sources
What each one is
InternVL3.5
Shanghai AI Lab's open vision-language family (1B to 241B, August 2025), strong on OCR and document benchmarks.
InternVL3.5 detailsPhi-4-multimodal
Microsoft's 5.6B text, vision and audio model (February 2025). Scores 84.4 on OCRBench per its model card.
Phi-4-multimodal detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "opengvlab/internvl","fallbacks": ["microsoft/phi-4-multimodal"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
InternVL3.5 alternativesPhi-4-multimodal alternativesBest OCR models in 2026