Mistral × New Model Releases

Mistral ships OCR 4 across 170 languages

Mistral ships OCR 4 across 170 languages

✎ Story body

Mistral released OCR 4, its document-analysis model — 170 languages, character-position output, per-character confidence scores. Practical multilingual document workflows just got a cleaner off-the-shelf option. Arxiv work reviewing transformer-based LMs across domain verticals ran alongside, keeping alive the debate over whether specialized models or generalist LMs win in vertical tasks. OCR 4 is a vote for specialized-and-fast. The story to watch: third-party accuracy audits on non-Latin scripts and actual enterprise adoption at scale.

▲ Official & Press
Official

Introducing Mistral OCR 4

Mistral AI News ・ 2026-06-23 ・ 📌

Mistral unveils Mistral OCR 4 with 170-language enterprise OCR

Press

Mistral、文書解析OCRの新版「OCR 4」公開 文字の位置や信頼度スコアを出力、日本語を含む170言語に対応

ITmedia AI+ ・ 2026-06-24

Mistral unveils OCR 4: 170 languages, bounding boxes, confidence scores

Academic (arxiv etc.) 3 ▾
Academic

Memory Depth, Not Memory Access: Selective Parametric Consolidation for Long-Running Language Agents

arXiv cs.LG (Machine Learning) ・ 2026-06-25

Academic

Steering Vision-Language Models with Joint Sparse Autoencoders

arXiv cs.AI (Artificial Intelligence) ・ 2026-06-24

Academic

Transformer-Based Language Models Across Domain Verticals: Architectures, Applications and Critical Assessment

arXiv cs.CL (Computation and Language) ・ 2026-06-23

← Story Archive