Two free AI tools for reading documents launched days apart. Only one is free for every business

Within days of each other in late October, two teams behind popular open document-AI projects shipped new OCR models, which turn scanned or photographed documents into usable text. Both are small enough to fit on one consumer graphics card. For any company building on them, the key difference is the license. MiniCPM-OCR is Apache 2.0, which allows commercial use, from code to weights. Surya 2.5's weights are free only for research, personal use and startups below $5 million in funding or revenue.

OpenBMB, announcing jointly with the Chinese AI startup ModelBest, released MiniCPM-OCR 1.0, a 1.3-billion-parameter document reader. It also released Jev, a 2-billion-parameter companion that grades OCR output instead of producing it. Datalab, the company behind the Surya OCR toolkit, shipped Surya 2.5, a 689-million-parameter model. Both are drawing a burst of developer attention on Hugging Face.

What actually shipped

MiniCPM-OCR 1.0 is the first OCR-focused model in the MiniCPM line. It targets Chinese and English documents, including text, tables, formulas, charts, seals and key-field extraction. The developers say it holds up on scanned, photographed, curved and tilted pages. All headline numbers are self-reported: 93.68 on Real5-OmniDocBench, 54.7 on OCRBench v2 and 82.7 overall on olmOCR-Bench.

OpenBMB reports about 1.2 pages per second on an RTX 5090 processing one image at a time. With its DSpark speed-up add-on, it says it reaches roughly 1.49 times the throughput of PaddleOCR-VL-1.6 and 3.62 times that of MinerU2.5 Pro. The card is loose with its own figures. Its header lists the model at 1B parameters while the body says 1.3B. Its metadata records 93.75 on Real5-OmniDocBench, not the 93.68 in the highlights.

Surya 2.5 takes a different approach. One small model handles page layout, reading full pages or individual blocks, and table recognition. Datalab's numbers, also self-reported, show steady gains over Surya 2 rather than a leap: 85.9 versus 83.3 on OmniParseBench, an open benchmark Datalab built itself; 83.7 versus 82.8 on olmOCR-bench; and 74.9% versus 74.5% on an internal set covering 91 languages.

The bigger change is speed. Datalab reports 6.19 pages per second on an H100 at 96 DPI, against 2.28 for Surya 2. At the 192 DPI setting it uses for its benchmarks, the figures are 4.52 and 2.24. Datalab also lists limits. Surya 2.5 is built for documents, not photos or natural scenes. It no longer outputs table row and column geometry, so users who need that must keep the older Surya 2 model.

A model that checks the other model's work

Jev targets a real weakness of document AI: OCR mistakes are usually silent. A dropped digit in a financial table looks just like a correct one, and the error travels downstream unnoticed. Jev reads nothing on its own. Given a page image and some OCR engine's output, it scores each text, formula and table region between 0 and 1. In OpenBMB's proposed workflow, high-scoring regions pass straight through and low-scoring ones go to a human reviewer. When several OCR engines run on the same page, Jev picks the best result.

OpenBMB reports an AUROC of 93.6 for Jev on JevOCRBench. AUROC measures how well a score separates good outputs from bad ones; 50 is a coin flip and 100 is perfect. That compares with 90.1 for JEV-35B-A3B, a 35-billion-parameter judge, and 93.8 for Qwen3.5-397B-A17B, a model roughly 200 times larger. The team also says Jev stays between 91.9 and 93.4 when judging four different OCR systems, including MiniCPM-OCR 1.0 itself. That supports its claim to work with any engine.

Some skepticism is warranted. The benchmark comes from the same team, and no third-party evaluation exists yet. The official press release says only that Jev's judgment accuracy is usually associated with far larger models, without giving figures. OpenBMB does publish a reproduction guide, so outside testing is possible.

Similar scores, very different licenses

On accuracy, the two are hard to separate. The one public benchmark both teams report is olmOCR-bench, and they land a point apart: 82.7 for MiniCPM-OCR and 83.7 for Surya 2.5. Each team ran its own evaluation under its own settings, and Surya's card compares only against Surya 2. So that gap is not a real head-to-head result. Cost, language coverage and licensing will decide most choices.

Licensing is where they split. OpenBMB's repository says the whole project, including code, documentation and weights, is under Apache 2.0, which permits commercial use with no revenue test. Datalab releases Surya's code under Apache 2.0 but puts the weights under a modified AI Pubs Open Rail-M license. That license is free for research, personal use and startups under the $5 million funding-or-revenue line. Anyone above it needs a commercial license from Datalab.

The terms fit Datalab's business: its model card points users to a paid managed platform and a pricing page for commercial licensing. They also mean a fast-growing startup building on Surya 2.5 must recheck its status after each funding round. A legal team will treat the two very differently, and calling both "open" hides that. AI Weekly framed Surya 2.5 as pricing pressure on closed document-AI services, and that pressure works differently depending on the license.

What the download spike does and doesn't mean

This week's momentum is real but starts from small numbers. Both models' monthly downloads were still in the thousands to low tens of thousands on Wednesday. Their pages showed 417 likes for MiniCPM-OCR and 34 for Surya 2.5. At that scale, big day-over-day percentage jumps mean developers are trying the models, not running them in production. Community activity is building faster around MiniCPM-OCR. Its page lists seven compressed variants, six versions retrained by others and a set of demos, including one comparing it directly against Baidu's Unlimited-OCR-2.

Three things are worth watching. First, whether anyone tests MiniCPM-OCR, Surya 2.5 and their peers on olmOCR-bench under one shared setup, which would settle the near-tie. Second, whether Jev's engine-agnostic scores hold up on messy real documents its authors did not choose. Third, whether Datalab's $5 million threshold costs it developers now that a permissively licensed model of similar size is a download away. Which small OCR models companies can actually ship is becoming a question about licenses as much as accuracy.