Gemini Flash vs SmolDocling
Pick Gemini Flash for bulk PDF to Markdown or JSON extraction, especially with mixed-language documents. It is a hosted API with a 1M-token context and a 92.62 OmniDocBench v1.6 score for Gemini 3 Flash. Pick SmolDocling only for English experiments on your own hardware. It is a 256M preview model that its authors say to replace with Granite-Docling.
Gemini Flashgoogle/gemini-flash | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| Type | Vision model | Open source |
| Price | 15 credits / page | 1 credit / page |
| Per 1,000 pages | $15.00 | $1.00 |
| Typical speed | 1.6 s / page | 1.0 s / page |
| Best for | Handwriting and mixed layouts | Very cheap page conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 100+ languages | Mainly English |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| 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 Gemini Flash if
- Gemini 3 Flash scores 92.62 on OmniDocBench v1.6, an independent benchmark, while SmolDocling has no published score there.
- You can send long documents in one request because the context window goes up to 1M tokens.
- It suits mixed-language documents and prompt or schema-driven JSON extraction, while SmolDocling is English only.
Choose SmolDocling if
- You can self-host it under Apache-2.0 per the model summary, so documents stay on your own machines.
- It has no per-token fee and runs on modest hardware at 256M parameters, at about 0.35 seconds per page on an A100 with vLLM.
- It returns DocTags, Markdown, HTML, JSON, LaTeX and bounding boxes, so you get location data without extra prompting.
Key differences
| Gemini Flash | SmolDocling | |
|---|---|---|
| How you run it | Hosted API through Gemini API or Vertex AI | Self-hosted, 256M parameters |
| Licence | Proprietary | Apache-2.0 per model summary (metadata tag says CDLA-Permissive-2.0) |
| Price | Gemini 3 Flash Preview: $0.50 per 1M input tokens, $3.00 per 1M output tokens; Batch/Flex $0.25 input and $1.50 output | Free, you pay for compute |
| Output formats | Markdown, JSON, HTML, plain text, shaped by your prompt with no fixed schema | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Benchmarks | 92.62 on OmniDocBench v1.6 (Gemini 3 Flash, independent) | Table TEDS 0.82 on FinTabNet, full-page OCR F1 0.80, OCRBench 338 (all from the authors) |
| Languages | Handles mixed-language documents | English only |
| Known failure | Cost per page varies with page density and output length; no confidence scores | Endless token repetition and output loops on some images, which make generation very slow |
Sources
- Gemini API pricing
- OmniDocBench leaderboard
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
What each one is
Gemini Flash
General vision model with strong handwriting and layout understanding at a low price.
Gemini Flash detailsSmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling 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": "google/gemini-flash","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Gemini Flash alternativesSmolDocling alternativesBest OCR models in 2026