| Mobile Web

Search results for vision language model

AI & Enterprise

Cohere unveils Parse 5 document-parsing AI focused on cost performance

Cohere has unveiled Parse 5, a document-parsing AI model that converts PDFs, slides and images into structured markdown, VentureBeat reported. The 2.3-billion-parameter vision-language model targets cost performance rather than maximum accuracy, the company said. On its ParseBench benchmark it scored 79.2 points, below GPT-5.5 and some other models but above several tools. Pricing is $1.50 per 1,000 pages, and a single-tenant platform is offered for high-volume users.

Industry

Dyna Robotics unveils DYNA-2 trained on 1 million hours of first-person human video

U.S. robotics company Dyna Robotics has unveiled DYNA-2, a robot-based model trained on about 1 million hours of first-person human video without using robot data. The company aims to address a shortage of training data for general-purpose robots by extracting action information from videos of people interacting with objects. It said success rates in precision manufacturing rose from about 20 percent to 80 to 90 percent and that fine-tuning for new platforms could be shortened.

Telecommunications & Media

Bamingtongwi opens 1,175,800 AI training datasets for broadcast video

The Broadcasting Media Communication Commission and the Korea Radio Promotion Association said on Aug. 5 they will open some AI training data for broadcast video for education and research through the National Information Society Agency’s AI-Hub secure zone. The release totals 1,175,800 items reflecting Korean culture and sentiment, spanning 3 areas and 10 types such as captioning, background and character generation, and production-environment upgrades. Access is free with prior application and approval.