Mistral OCR 4 Launches With Bounding Boxes, 170 Language Support at $2-4 Per 1,000 Pages
Mistral AI released OCR 4, a compact document extraction model that returns bounding boxes, block classification, and inline confidence scores alongside text. The model supports 170 languages, scores 85.20 on OlmOCRBench, and is priced at $4 per 1,000 pages via API ($2 with batch discount) or $5 per 1,000 pages through Document AI.
Mistral OCR 4 Launches With Bounding Boxes, 170 Language Support at $2-4 Per 1,000 Pages
Mistral AI released OCR 4, a document extraction model that adds bounding boxes, block classification, and inline confidence scores to extracted text. The model runs in a single container for self-hosted deployment and supports 170 languages across 10 language groups.
Pricing and Deployment
OCR 4 is priced at $4 per 1,000 pages via API, with a 50% batch discount reducing the cost to $2 per 1,000 pages. The Document AI interface in Mistral Studio costs $5 per 1,000 pages. The model is compact enough to deploy on a single container for organizations with data sovereignty requirements.
Performance Claims
Mistral claims OCR 4 achieves an 85.20 score on OlmOCRBench, the top result among models tested by the company. The model also scores 93.07 on OmniDocBench and 0.98 on Mistral's internal Crawl Multilingual evaluation.
In human preference evaluations, independent annotators preferred OCR 4 over competing systems with an average win rate of 72%, according to Mistral. The company tested OCR 4 against 600+ documents across 12+ languages in blind comparisons.
Mistral notes significant limitations in automated benchmark scoring, including ground-truth errors in reference data, equivalent LaTeX notation scored as mismatches, and multi-column reading order artifacts. The company recommends evaluating the model on your own documents rather than relying solely on benchmark scores.
Technical Capabilities
OCR 4 returns structured document representations with:
- Bounding boxes for text localization and in-context highlighting
- Block classification identifying titles, tables, equations, signatures, and other document elements
- Inline confidence scores per-page and per-word for verification workflows
- Format support for PDF, DOC, PPT, and OpenDocument files
- 170 languages across English, Western Europe, Eastern Europe, Middle Eastern, Chinese, East Asian, Southeast Asian, and rare language groups
The model integrates with Mistral Search Toolkit, an open-source search framework announced at the AI Now Summit, providing structured inputs for RAG and enterprise search pipelines.
Use Cases and Performance Data
Aidan Donohue, AI Engineer at Rogo, reported OCR 4 matched the accuracy of leading agentic document parsers on financial QA datasets at roughly 8x lower cost and 17x lower latency. Ivan Mihailov, AI engineer at Anaqua, stated the model is roughly 4x faster per page than their previous provider for high-volume docketing workflows.
What This Means
OCR 4 addresses a critical gap in document processing by combining text extraction with spatial and structural metadata. Bounding boxes enable citation-grounded outputs and data pipeline validation, while block classification supports semantic chunking for retrieval systems. The $2-4 per 1,000 page pricing undercuts many enterprise document services, though organizations should verify performance on their specific document types given the benchmark limitations Mistral acknowledges. The single-container deployment option makes this model accessible to organizations that cannot send documents to external APIs for compliance or sovereignty reasons.
Related Articles
Mistral Releases Shieldstral, a 3B Open-Weights Safety Classifier That Matches Models 7x Its Size
Mistral has released Shieldstral, a 3B open-weights safety classifier that reframes content moderation as a policy-adaptive question-answering task. The model claims to match or outperform guard models up to 7x its size on text safety and multimodal benchmarks, and runs on a single 16GB GPU.
Mistral's 3B-Parameter Shieldstral Matches 20B Safety Model on Text Benchmarks
Mistral's new Shieldstral, a 3-billion-parameter open-weight safety classifier, posts an 84.9% F1 score on text benchmarks—tying OpenAI's GPT-OSS-Safeguard-20B, a model roughly seven times larger. The model lets operators define safety rules at runtime using plain-language yes/no questions instead of fixed taxonomies.
Mistral AI Releases Shieldstral-1.0-3B, a 3B-Parameter Policy-Adaptive Safety Classifier
Mistral AI has released Shieldstral-1.0-3B, a compact open-weight safety classifier that evaluates text and images against natural-language policies specified at inference time. The 3B model runs on a single GPU and reports F1 scores competitive with or exceeding larger moderation models like LlamaGuard-4-12B and GPT-OSS-Safeguard-20B on multiple benchmarks.
OpenAI Halts Parts of Astra Model Development After It Hit 'Critical' Cybersecurity Threshold
OpenAI disclosed that its in-development Astra model showed cyberattack capabilities strong enough that it cannot rule out a 'Critical' risk classification. The company has paused related internal activity and added security controls under its Preparedness Framework.
Comments
Loading...