@VikParuchuri: OlmOCR is the best designed OCR benchmark, but it's saturated. Gains are now mostly from postprocessing to match tests,…
Summary
Vik Paruchuri argues that OlmOCR is the best-designed OCR benchmark but has become saturated, with recent gains stemming mainly from postprocessing tricks to match test sets rather than genuine model improvements.
View Cached Full Text
Cached at: 10/11/26, 01:10 PM
OlmOCR is the best designed OCR benchmark, but it’s saturated.
Gains are now mostly from postprocessing to match tests, not from model quality improvement. https://t.co/mKdh62y8zo
Similar Articles
@VikParuchuri: Marker is powered by the new Surya OCR 2 model (https://github.com/datalab-to/surya…), which is the highest-scoring mod…
Datalab announces Surya OCR 2, a 650M parameter OCR model that achieves top accuracy under 1B parameters on olmOCR benchmarks, with high speed and multilingual support.
We benchmarked 18 LLMs on OCR (7k+ calls) — cheaper/old models oftentimes win. Full dataset + framework open-sourced. [R]
A comprehensive benchmark of 18 LLMs on OCR tasks (7k+ calls) reveals that cheaper and older models often match premium accuracy at a fraction of the cost, with full dataset and framework open-sourced.
@VikParuchuri: Surya 2, which has 650M params and scores 83.3% on olmocr, is the most accurate small OCR model. One reason why is char…
Surya 2 is a 650M parameter OCR model achieving 83.3% on olmocr, claiming to be the most accurate small OCR model thanks to character tokenization which improves accuracy and model size.
@techNmak: A lightweight VLM that beats the giants at OCR. (1.7B parameters, SOTA on OmniDocBench) dots. ocr is a new multilingual…
dots.ocr is a new lightweight 1.7B parameter multilingual vision-language model that achieves state-of-the-art performance on OmniDocBench, outperforming much larger models (72B+) at document parsing and OCR tasks.
Newer Models, Same Advantage
DharmaOCR, a model specialized for Brazilian Portuguese OCR, outperforms newer models like Mistral OCR4 through domain-specific fine-tuning and Direct Preference Optimization. The article explains the training pipeline and presents benchmark results.