AI BriefingDharmaOCR releases benchmarks
·2026.04.30 01:08
Key point
DharmaOCR released specialized 3B and 7B SLMs along with cost and performance benchmarks.
Details
Released DharmaOCR on Hugging Face, and also made all related models and datasets open.
- Fine-tuned open-source 3B and 7B SLMs using SFT + DPO.
- The comparison targets were GPT-5.4, Gemini 3.1 Pro, Claude Opus 4.6, Google Document AI, and the open-source alternatives OlmOCR, Deepseek-OCR, GLMOCR, and Qwen3.
- The specialized models recorded the highest performance, scoring 7B 0.925 and 3B 0.911.
- The DPO approach, which uses the model's degenerate outputs as rejected examples, reduced the failure rate by 87.6%.
- AWQ quantization cut inference cost per page by about 22%, with almost no performance degradation.
A paper was also released alongside this, where the experimental methods and comparison results can be found.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis PerformancePrivatemode Benchmarks GLM-5.3-Flash as Vision-Capable Decision Model with Parity to Jev but Higher Cost