{"id":13785,"date":"2026-11-17T11:02:42","date_gmt":"2026-11-17T02:02:42","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/?p=13785"},"modified":"2026-11-17T11:11:51","modified_gmt":"2026-11-17T02:11:51","slug":"openbmb-releases-quantized-minicpm-ocr-jev","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/openbmb-releases-quantized-minicpm-ocr-jev\/","title":{"rendered":"OpenBMB Releases Quantized Versions of MiniCPM-OCR-Jev"},"content":{"rendered":"<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-GPTQ\">openbmb\/MiniCPM-OCR-Jev-GPTQ<\/a><\/td>\n<\/tr>\n<tr>\n<td>Family guide<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/model-openbmb-minicpm-ocr-jev-en\/\">MiniCPM-OCR-Jev guide (1 articles)<\/a><\/td>\n<\/tr>\n<tr>\n<td>Publisher guide<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/publisher-openbmb-en\/\">OpenBMB: models and licenses<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-10-26<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>apache-2.0<\/td>\n<\/tr>\n<tr>\n<td>Formats<\/td>\n<td>safetensors<\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p>openbmb has released quantized versions (in GPTQ and AWQ formats) of &#8220;MiniCPM-OCR-Jev&#8221;, a model designed for OCR quality evaluation. Given text, formula, or table region images along with candidate OCR results, this model returns a score from 0 to 1 indicating the likelihood that the candidate is correct.<\/p>\n<p>This model is designed to automatically accept reliable regions, route uncertain ones to human review, or select the optimal result among multiple OCR models.<\/p>\n<h2>Specifications<\/h2>\n<p>Specifications for the source model, MiniCPM-OCR-Jev, are as follows:<\/p>\n<ul>\n<li>Number of parameters: 2B<\/li>\n<li>Architecture: Based on Qwen3.5-2B, with the language model output head replaced by a binary classification head<\/li>\n<\/ul>\n<h2>Performance<\/h2>\n<p>The released quantized versions inherit the performance of the source MiniCPM-OCR-Jev model. According to evaluations in the model card, the decision AUROC (\u00d7100) on JevOCRBench is 93.6. Despite having 2B parameters, it outperforms JEV-35B-A3B (90.1) and approaches the performance of Qwen3.5-397B-A17B (93.8).<\/p>\n<p>It also demonstrates stable versatility against analysis results from other OCR models, with AUROC (\u00d7100) ranging between 91.9 and 93.4 when using analysis results from MiniCPM-OCR 1.0, GLM-OCR, PaddleOCR-VL-1.6, and Xiaomi-OCR-0.<\/p>\n<p>A comparison of discrimination performance for each OCR model, as stated in the model card, is as follows:<\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<p class=\"lmw-table-hint\" style=\"margin:0 0 4px;font-size:0.85em;opacity:0.7;\">\u2192 Scroll horizontally to see all columns<\/p>\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th style=\"text-align: left;\">OCR model<\/th>\n<th style=\"text-align: right;\">MiniCPM-OCR 1.0<\/th>\n<th style=\"text-align: right;\">GLM-OCR<\/th>\n<th style=\"text-align: right;\">PaddleOCR-VL-1.6<\/th>\n<th style=\"text-align: right;\">Xiaomi-OCR-0<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td style=\"text-align: left;\">AUROC (\u00d7100)<\/td>\n<td style=\"text-align: right;\">92.7<\/td>\n<td style=\"text-align: right;\">91.9<\/td>\n<td style=\"text-align: right;\">93.4<\/td>\n<td style=\"text-align: right;\">93.1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p>This table shows that MiniCPM-OCR-Jev can accurately distinguish the correctness of results generated by existing OCR models such as MiniCPM-OCR 1.0, PaddleOCR-VL-1.6, and Xiaomi-OCR-0.<\/p>\n<h2>Strengths and Use Cases<\/h2>\n<p>MiniCPM-OCR-Jev functions as a powerful tool to ensure the reliability of OCR processing in document analysis. It proves particularly valuable in quality control when automatically digitizing large volumes of documents.<\/p>\n<p>Specific operational scenarios include the following three:<br \/>\n&#8211; <strong>Automation and efficiency in OCR workflows<\/strong>: Based on the scores calculated by this model, you can build a workflow where highly reliable results are directly fed into the system, while only &#8220;suspicious&#8221; low-score sections are extracted and presented to humans. This maintains high analysis accuracy while eliminating the effort of human verification for every single item.<br \/>\n&#8211; <strong>Accuracy improvement through ensemble methods<\/strong>: Run multiple OCR engines (such as MiniCPM-OCR 1.0, PaddleOCR-VL-1.6, and Xiaomi-OCR-0) on the same image region, and evaluate each output with this model. By adopting the candidate that achieves the highest score as the final result, you can obtain more accurate analysis results even for complex formulas, tables, or text with special fonts that are prone to misreading by a single model.<br \/>\n&#8211; <strong>Support for diverse regions<\/strong>: A major strength is its ability to determine consistency with images with high precision not only for text, but also for regions traditionally difficult to evaluate with standard OCR, such as formulas and tables.<\/p>\n<p>When integrating into a pipeline, the configuration sends document region images, categories, and text candidates output by OCR engines to the model&#8217;s inference service. The service performs internal rendering and returns an image-text alignment score. Developers can branch subsequent processing dynamically by comparing this score against a threshold.<\/p>\n<p>For API usage, endpoints are available such as <code>score<\/code> for evaluating a single candidate, <code>score_best<\/code> for selecting the optimal one from multiple candidates, and <code>score_best_batch<\/code> for processing multiple blocks together. Additionally, when using inference results, it is recommended to verify not just the numerical score, but also that <code>verdict<\/code> (the judgment status) is set to <code>scored<\/code>. This prevents errors such as rendering failures or pre-check errors from being mistakenly treated as low scores.<\/p>\n<p><!-- lmw:hardware --><\/p>\n<h2>Hardware Requirements<\/h2>\n<p><strong>Estimated requirements (calculated by Local Model Watch)<\/strong> \u2014 2.2B parameters (taken from the base model openbmb\/MiniCPM-OCR-Jev)<\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Your VRAM<\/th>\n<th>Quantization<\/th>\n<th>File size<\/th>\n<th>Est. memory needed<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>4GB (laptop iGPU \/ phone class)<\/td>\n<td>GPTQ 4bit<\/td>\n<td>2.2GB<\/td>\n<td>2.7GB<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><strong>Inference engine support<\/strong> (architecture name matched against each project&#8217;s own model registry in its source code, checked 2026-11-17): llama.cpp: not registered, vLLM: not registered, MLX (mlx-lm): registered. &#8220;Not registered&#8221; means the name is absent from that registry today, not that the model cannot run.<\/p>\n<p><em>Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model&#8217;s authors. Compare with other models in our <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/\">VRAM quick reference<\/a>. What the quantization names mean: <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/glossary-quantization-en\/\">glossary<\/a>.<\/em><\/p>\n<p><!-- \/lmw:hardware --><\/p>\n<p><!-- lmw:runnability --><\/p>\n<h2>Can You Run It Locally?<\/h2>\n<p><strong>Runs in Ollama, LM Studio and llama.cpp via a converted build.<\/strong><\/p>\n<p>The publisher ships safetensors, but <a href=\"https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-gguf\">openbmb\/MiniCPM-OCR-Jev-gguf<\/a> provides a GGUF build you can use.<\/p>\n<p><strong>License \u2014 <code>apache-2.0<\/code> (Commercial use allowed):<\/strong> Permits commercial use, modification and redistribution. Redistribution requires including the license and stating changes; includes a patent grant.<\/p>\n<p><em>Compiled by this site&#8217;s code from the published formats, converted builds we have found, and each engine&#8217;s own model registry. &#8220;Not found&#8221; means we have not seen such a build, not that none exists. License summaries are not legal advice \u2014 check the publisher&#8217;s original terms before relying on them.<\/em><\/p>\n<p><!-- \/lmw:runnability --><\/p>\n<p><!-- lmw:peers --><\/p>\n<h2>Recent Models in the Same Size Class<\/h2>\n<p><em>Models with <\/em><em>up to 4B<\/em><em> parameters that Local Model Watch covered recently, listed by code from our article log for comparison. VRAM tiers are this site&#8217;s estimates; licenses are as stated on the model cards.<\/em><\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<p class=\"lmw-table-hint\" style=\"margin:0 0 4px;font-size:0.85em;opacity:0.7;\">\u2192 Scroll horizontally to see all columns<\/p>\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Model<\/th>\n<th>Parameters<\/th>\n<th>Smallest VRAM tier<\/th>\n<th>License<\/th>\n<th>Our article<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>LiquidAI\/d1-omni-600M<\/td>\n<td>587M<\/td>\n<td>4GB<\/td>\n<td>lfm1.0<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/liquidai-d1-omni-600m-multimodal-decision-model\/\">d1-omni-600M Multimodal Decision Model: 4GB+ VRAM, GGUF Builds<\/a> (2026-11-17)<\/td>\n<\/tr>\n<tr>\n<td>LiquidAI\/d1-3B<\/td>\n<td>3.1B<\/td>\n<td>4GB<\/td>\n<td>lfm1.0<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/d1-3b-decision-model\/\">d1-3B Vision-Language Model: 4GB+ VRAM, GGUF Builds<\/a> (2026-11-17)<\/td>\n<\/tr>\n<tr>\n<td>baidu\/Unlimited-OCR-2<\/td>\n<td>3.3B<\/td>\n<td>8GB<\/td>\n<td>mit<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/baidus-unlimited-ocr-2-for-long-documents\/\">Unlimited-OCR-2 Vision-Language Model: 8GB+ VRAM<\/a> (2026-11-17)<\/td>\n<\/tr>\n<tr>\n<td>openbmb\/MiniCPM-OCR-1.0-DSpark<\/td>\n<td>297M<\/td>\n<td>4GB<\/td>\n<td>apache-2.0<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/minicpm-ocr-1-0-dspark-overview\/\">MiniCPM-OCR-1.0-DSpark OCR Acceleration Draft Model: 4GB+ VRAM<\/a> (2026-11-17)<\/td>\n<\/tr>\n<tr>\n<td>LiquidAI\/d1-2B-GGUF<\/td>\n<td>1.9B<\/td>\n<td>4GB<\/td>\n<td>lfm1.0<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/liquidai-d1-2b-gguf-2\/\">d1-2B-GGUF Text, Image, and Video Decision-Making Model: 4GB+ VRAM<\/a> (2026-11-17)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><!-- \/lmw:peers --><\/p>\n<h2>How to Get It<\/h2>\n<p>The newly released models are 4-bit quantized (W4A16) in GPTQ and AWQ formats. They are distributed in safetensors format under the Apache-2.0 license. Hugging Face gating (agreement to terms of use) is not required, and anyone can download and start using them without restrictions.<\/p>\n<p>vLLM is supported as an inference engine, and you can spin up an API server loading the model locally or on a server by running the following commands:<\/p>\n<p>Startup command for the GPTQ version:<\/p>\n<pre><code class=\"language-bash\">vllm serve openbmb\/MiniCPM-OCR-Jev-GPTQ \\\n  --served-model-name MiniCPM-OCR-Jev \\\n  --trust-remote-code\n<\/code><\/pre>\n<p>Startup command for the AWQ version:<\/p>\n<pre><code class=\"language-bash\">vllm serve openbmb\/MiniCPM-OCR-Jev-AWQ \\\n  --served-model-name MiniCPM-OCR-Jev \\\n  --trust-remote-code\n<\/code><\/pre>\n<p>Additionally, the &#8220;Quick start&#8221; section of the model card provides links to deployment guides for various other backends such as SGLang, llama.cpp (GGUF format), Ollama, and FlagOS. A system is in place allowing users to select and deploy the optimal format for their infrastructure environment and preferred inference engine.<\/p>\n<p><!-- lmw:variants --><\/p>\n<h2>Quantized and Converted Variants<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<p class=\"lmw-table-hint\" style=\"margin:0 0 4px;font-size:0.85em;opacity:0.7;\">\u2192 Scroll horizontally to see all columns<\/p>\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Added<\/th>\n<th>Publisher<\/th>\n<th>Format<\/th>\n<th>Repository<\/th>\n<th>Smallest VRAM tier (build, est. memory)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>2026-10-28<\/td>\n<td>openbmb<\/td>\n<td>GGUF<\/td>\n<td><a href=\"https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-gguf\">openbmb\/MiniCPM-OCR-Jev-gguf<\/a><\/td>\n<td>Q8_0 2.2GB (fits in 4GB VRAM)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p>File sizes of each build:<\/p>\n<ul>\n<li>Available builds in openbmb\/MiniCPM-OCR-Jev-gguf: IQ3_XS 1.0GB \/ IQ3_M 1.0GB \/ IQ4_XS 1.1GB \/ Q4_0 1.1GB \/ Q4_K_S 1.1GB \/ IQ4_NL 1.2GB \/ Q4_K_M 1.2GB \/ Q4_1 1.2GB \/ Q5_0 1.3GB \/ Q5_K_S 1.3GB \/ Q5_K_M 1.3GB \/ Q5_1 1.4GB \/ Q6_K 1.4GB \/ Q8_0 1.9GB \/ F16 3.5GB<\/li>\n<\/ul>\n<p><em>This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files. What the quantization names mean: <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/glossary-quantization-en\/\">glossary<\/a>.<\/em><\/p>\n<p><!-- \/lmw:variants --><\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/minicpm-ocr-1-0-dspark-overview\/\">MiniCPM-OCR-1.0-DSpark OCR Acceleration Draft Model: 4GB+ VRAM<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/minicpm-v-4-7-1b-dspark-gguf-released\/\">MiniCPM-V-4.7-1B-DSpark-gguf Text Generation Model: 4GB+ VRAM<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/minicpm-v-4-7-35b-a3b-dspark-gguf-2\/\">MiniCPM-V-4.7-35B-A3B-DSpark-gguf Draft Model for Speedup: 4GB+ VRAM<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/09\/08\/minicpm5-2b-released\/\">MiniCPM5-2B Text Generation Model: Our Test Answers, 4GB+ VRAM<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Find models by VRAM<\/strong> (This model runs from the 4GB tier) \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-8gb-en\/\">Other models that run on a 8GB GPU<\/a><\/li>\n<li><strong>Explore the same model family<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/model-openbmb-minicpm-ocr-jev-en\/\">MiniCPM-OCR-Jev family overview (1 articles, 1 converted builds)<\/a><\/li>\n<li><strong>Engines that run this model<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-llama-cpp-en\/\">llama.cpp<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-ollama-en\/\">Ollama<\/a><\/li>\n<li><strong>Formats this model is available in<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-gguf-en\/\">GGUF format guide and models<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-gptq-en\/\">GPTQ format guide and models<\/a><\/li>\n<li><strong>Learn about the publisher<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/publisher-openbmb-en\/\">OpenBMB: models, licenses and articles<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-GPTQ\">https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-GPTQ<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-AWQ\">https:\/\/huggingface.co\/openbmb\/MiniCPM-OCR-Jev-AWQ<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Explore the GPTQ and AWQ quantized versions of MiniCPM-OCR-Jev for OCR quality evaluation, featuring hardware requirements, benchmarks, and usage.<\/p>\n","protected":false},"author":1,"featured_media":13784,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[310],"tags":[4213,4215,4217,1492,320,1547,592,1952],"class_list":["post-13785","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-new-models","tag-awq-en","tag-gptq-en","tag-minicpm-ocr-jev-en","tag-ocr-en","tag-openbmb-en","tag-verified","tag-vllm-en","tag-vlm-en"],"lang":"en","translations":{"en":13785,"ja":13783},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/13785","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=13785"}],"version-history":[{"count":2,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/13785\/revisions"}],"predecessor-version":[{"id":17962,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/13785\/revisions\/17962"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/13784"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=13785"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=13785"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=13785"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}