AIvengo > Reviews > New model from DeepSeek recognizes documents cheaply and efficiently

New model from DeepSeek recognizes documents cheaply and efficiently

DeepSeek rolled out a new model for document recognition. And you know what? It doesn’t just read text from pages – it understands structure. And does this cheaply and efficiently, which is rare in the AI world.

This wonder is called DeepSeek-OCR, and the difference from classic optical character recognition systems is fundamental. Regular OCR simply extract text. But this model immediately restores document structure: headings, lists, tables, figure captions. Outputs result in Markdown format, which is convenient for indexing and subsequent work of neural networks.

The main feature – so-called optical context compression. The model doesn’t retell every detail from the page, but squeezes out only what’s needed: text and semantic structure. This reduces data volume 20-fold. And fewer tokens – cheaper and faster processing by any subsequent language model.

DeepSeek-OCR uses visual tokens. These are conditional glances at parts of the image. Even with a small budget of 100 tokens, recognition accuracy holds at 97%. If the page is too complex, Gundam mode turns on. The document is automatically divided into fragments, and difficult areas are analyzed separately without speed loss.

In benchmarks the system showed impressive results. And accuracy practically doesn’t drop even with minimal number of visual tokens, and compression ratio reaches 20-fold. Efficiency in pure form.

Autor: AIvengo

For 5 years I have been working with machine learning and artificial intelligence. And this field never ceases to amaze, inspire and interest me.

Scientists became more afraid of AI hallucinations

The more scientists work with artificial intelligence, the less they trust it. Academic publisher Wiley released a preliminary report for 2025 on technology's impact on science, and the conclusions there are paradoxical. Researchers began treating neural networks with greater skepticism than a year ago, when the technology was obviously less developed.

New model from DeepSeek recognizes documents cheaply and efficiently

DeepSeek rolled out a new model for document recognition. And you know what? It doesn't just read text from pages - it understands structure. And does this cheaply and efficiently, which is rare in the AI world.

OpenAI officially denied GPT-6 release by end of year

At OpenAI they decided to cool public expectations and confessed: GPT-6 won't happen this year. But don't rush to be upset - this doesn't mean the company is sitting idle.

Father of reinforcement learning predicted end of large language models era

Richard Sutton - this is one of the fathers of reinforcement learning and Turing Award laureate. So he stated that the era of large language models is coming to an end. Next, in his opinion, comes the era of experience. And here's why he thinks so.

Artificial intelligence detects ADHD without questionnaires and doctors

Imagine you could diagnose ADHD simply by how your brain processes flickering letters on a screen. No questionnaires, no months of waiting for an appointment with a specialist. AI looks at your visual rhythms and gives a verdict with 92% accuracy. Sounds like science fiction? But this is already reality.