Russia Develops a Universal AI Model for Handling Any Type of Document
The system, built by MWS AI—a subsidiary of MTS Web Services—can analyze both text and images simultaneously.
Russian developer MWS AI (part of MTS Web Services) has unveiled its first multimodal AI model, Cotype VL, along with a benchmarking tool called MWS Vision Bench. The presentation took place at the Finopolis 2025 forum, according to Forbes.
Built on an open architecture, Cotype VL contains 32 billion parameters, enabling it to recognize printed and handwritten text within images, interpret context for accurate translation, describe visuals, and answer complex questions about image content. The model supports Russian, English, and Chinese. Its standout capability is processing complex document types—such as diagrams, blueprints, spreadsheets, and scanned contracts.
A New Benchmark for Russia’s AI Ecosystem
MWS Vision Bench is the first open benchmark designed to evaluate multimodal AI systems focused on Russian-language business documents. It includes 2,580 tasks and 800 images, providing a consistent basis for comparing models that handle real-world corporate and government document workflows.
The launch of Cotype VL and MWS Vision Bench marks a turning point: multimodal AI is moving from experimental phases to mainstream commercialization. By introducing its own benchmark, Russia’s AI industry has closed a methodological gap—enabling domestic developers to test and validate models using native business data.
