MiniCPM-OCR 1.0 vs Unlimited-OCR-2

Official MiniCPM-OCR 1.0 efficiency and quality figure from the OpenBMB GitHub assets — OpenBMB/MiniCPM-OCR assets/minicpm-ocr/efficiency-performance-en.png. Source · License: Apache-2.0.

Official Baidu Unlimited-OCR-2 card figure — baidu/Unlimited-OCR-2 assets/Unlimited-OCR2.png. Source · License: Baidu card image.

Verdict. Pick MiniCPM-OCR 1.0 if you want OpenBMB’s ungated Apache-2.0 ~1.3B OCR-specific MiniCPM-V/o card (zh/en; GitHub LICENSE file; SDK + layout model named). Pick Unlimited-OCR-2 if you want Baidu’s official ungated ~3.3B long-document card that names SR-SWA and a 131,072 config context (MIT tag; no LICENSE blob on 2026-10-28). Both are official org cards. Vendor benches stay vendor. We have no sourced bake-off.

Compare MiniCPM-OCR 1.0 and Unlimited-OCR-2.

Dimension	MiniCPM-OCR 1.0	Unlimited-OCR-2	Source
Developer / host	OpenBMB (huggingface.co/openbmb/MiniCPM-OCR-1.0)	Baidu (huggingface.co/baidu/Unlimited-OCR-2)	Hugging Face API
License	apache-2.0 card tag; GitHub LICENSE Apache-2.0; HF LICENSE 404	mit card tag; LICENSE sibling 404 (2026-10-28)	HF cardData / GitHub LICENSE (2026-10-29 / 2026-10-28)
Gating	No	No	Hugging Face API
Parameters (safetensors)	1,300,434,160 (~1.30B)	3,343,978,497 (~3.34B)	Hugging Face API
Architecture	minicpm_ocr / MiniCPMOcrForConditionalGeneration	unlimited-ocr / UnlimitedOCRForCausalLM	config.json
Context (config.json)	262,144 (text_config.max_position_embeddings)	131,072	config.json
Languages (card)	Chinese and English	multilingual (card)	Official READMEs
Hugging Face likes / downloads	261 / 791 (2026-10-29)	228 / 513 (2026-10-28)	Hugging Face API
GitHub	OpenBMB/MiniCPM-OCR · 237★ / 16 forks · Apache-2.0	No first-party repo opened on the Unlimited-OCR-2 page	GitHub API (2026-10-29)
Quality vs Acrobat OCR	Not independently tested	Not independently tested	n/a

When to choose which

Compact ~1.3B Apache-2.0 OCR with a GitHub LICENSE: MiniCPM-OCR 1.0 — Identification of the card, not a quality rank.
Baidu official long-document card with SR-SWA named: Unlimited-OCR-2 — MIT tag; no LICENSE blob on 2026-10-28. Not a quality rank.
Same-org MiniCPM chat VLM: neither — That is MiniCPM-V 4.7, a different job.
DeepSeek optical-compression OCR card: neither — See DeepSeek-OCR and Unlimited-OCR-2 vs DeepSeek-OCR.

Read the entity pages: MiniCPM-OCR 1.0 and Unlimited-OCR-2. Both hang off Adobe Acrobat alternatives. MiniCPM-OCR 1.0 is not MiniCPM-V 4.7. Unlimited-OCR-2 is Baidu’s official card, not a random fine-tune. DeepSeek’s OCR card is a third entity: DeepSeek-OCR — Unlimited-OCR-2 vs DeepSeek-OCR.

Why this comparison exists

MiniCPM-OCR 1.0 (card created 2026-10-21) is OpenBMB’s OCR-specific ~1.3B card. Unlimited-OCR-2 is already on this site as Baidu’s ~3.3B long-document card. Same job class (document OCR weights). Different size, license file, and family. This is identification, not a bake-off.

Sourced snapshot (2026-10-29)

Field	MiniCPM-OCR 1.0	Unlimited-OCR-2
Host	huggingface.co/openbmb/MiniCPM-OCR-1.0 (not gated)	huggingface.co/baidu/Unlimited-OCR-2 (not gated)
HF likes / downloads	261 / 791	228 / 513 (2026-10-28)
License	apache-2.0 tag + GitHub LICENSE; HF blob missing	mit tag; no LICENSE file (2026-10-28)
Size (tensors)	1,300,434,160	3,343,978,497
config.json context	262,144 (`text_config`)	131,072
Job	image-text-to-text (document OCR)	image-text-to-text (document OCR)
Quality vs Acrobat	Not independently tested	Not independently tested

License

MiniCPM-OCR 1.0: Hub license:apache-2.0; GitHub Apache License 2.0; Hugging Face LICENSE blob missing on 2026-10-29. Unlimited-OCR-2: Hub license:mit; LICENSE blob missing on 2026-10-28. Not legal advice.

How to choose

Need OpenBMB’s ~1.3B OCR card (Apache-2.0 GitHub LICENSE): MiniCPM-OCR 1.0.
Need Baidu’s official long-doc card (SR-SWA named): Unlimited-OCR-2.
Need DeepSeek’s MIT LICENSE + GitHub repo: DeepSeek-OCR.
Need a PDF toolbox, not OCR weights: Stirling-PDF — Stirling-PDF vs Adobe Acrobat.
Need MiniCPM chat VLM, not OCR: MiniCPM-V 4.7.

See also Writing & Office. Catalog: home. Editorial: About.

FAQ

Which is more accurate on long PDFs?

We do not rank them. MiniCPM-OCR’s OmniDocBench / Real5 / pages/s lines stay vendor. Unlimited-OCR-2’s MPDocBench / MarathonDocBench lines stay vendor. Do not merge those tables.

Is MiniCPM-OCR a fine-tune of Unlimited-OCR-2 or DeepSeek-OCR?

The MiniCPM card says it is trained for OCR on MiniCPM-V 4.6. Treat it as OpenBMB’s official OCR card. Unlimited-OCR-2 is Baidu’s official card. DeepSeek-OCR is DeepSeek’s.

Why this pairing instead of DeepSeek-OCR?

Unlimited-OCR-2 is already compared with DeepSeek-OCR. MiniCPM-OCR 1.0 is a new same-site OCR entity; the split here is size, license file, and family (MiniCPM-V/o vs Baidu SR-SWA).

Can either replace Acrobat Pro OCR?

Not independently tested. PDF toolboxes stay on Adobe Acrobat alternatives (Stirling-PDF, PDF Arranger, Okular).

Sources & verification

Last verified 2026-10-29 · Last updated 2026-10-29· Edited by forkalt editors · Editorial policy

License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.