PaddleOCR Releases

This release addresses critical stability issues, including a crash prevention fix in speaker diarization and bug fixes for NeMo transducer beam search. It also adds configuration flexibility for sense-voice audio processing.

May 8, 2026

A minor version bump primarily consisting of dependency updates, specifically upgrading the OpenTelemetry SDK to a newer version.

May 1, 2026

A security and bug fix release for the Fiber web framework that ensures HTML output is properly escaped in the context formatting method to prevent potential injection issues.

Apr 25, 2026

Deeply integrated with the Hugging Face ecosystem, added document format conversion tools, and released PaddleOCR.js for browser-based inference.

Apr 21, 2026

Added llama-cpp-server backend support, expanded hardware support to AMD and Intel Arc GPUs, fixed dependency issues for NPU and XPU, and removed page limits in Docker Compose.

Apr 14, 2026

Version bump to 0.2.87 includes a specific fix to pin an unpinned GitHub action.

Apr 1, 2026

Addresses stability issues in the web app, specifically fixing functionality for users who disabled memories without a chat model and correcting deprecation banners.

Mar 26, 2026

Focuses on compatibility updates to ensure web app page templates work correctly with Starlette 1.0.0.

Mar 25, 2026

Major performance improvements via Next.js 15 upgrade and a series of bug fixes addressing research agents, PDF processing, and memory leaks.

Mar 25, 2026

Version bump to 0.2.86 with dependency updates to address Node.js 20 deprecation warnings.

Mar 25, 2026

Version bump to 0.2.85 accompanied by dependency updates for OpenTelemetry SDK, Cloudflare Circl, and Go-git.

Mar 23, 2026

This beta release focuses on enhancing the Obsidian integration and user experience on the web app. Key improvements include better batch sync capabilities, autocomplete filters for queries, and the ability to copy references directly from the interface.

Feb 22, 2026

This update improves PDF parsing capabilities, specifically regarding table extraction and formatting. It also updates dependencies and adds support for different content types in the Accept header.

Feb 20, 2026

PaddleOCR-VL-1.5 complex document parsing solution released, achieving 94.5% accuracy on OmniDocBench v1.5 and surpassing top global general-purpose large models. The model innovatively supports irregular-shaped bounding box localization for document elements in real-world scenarios.

Jan 29, 2026

PaddleOCR-VL now supports custom model names and API keys for seamless integration with third-party inference services. PP-StructureV3 MCP Server adds support for Qianfan platform, and Muxi GPU inference support is added.

Jan 20, 2026

This beta release improves the web application by adding support for rendering mermaid diagrams in markdown code blocks and addresses several bugs including subscription failures and issues with AI file interactions in the code sandbox.

Jan 2, 2026

A maintenance release focused on security, bumping the `mammoth` and `pdfminer.six` dependencies to patch specific CVEs and security advisories.

Dec 1, 2025

Minor patch release with bug fixes and improvements. No significant new features documented in this release.

Nov 13, 2025

Bug fix release addressing issues with document image preprocessing switch in PP-StructureV3 and PaddleOCR-VL. Documentation improved with offline environment usage instructions.

Oct 29, 2025

Major release introducing PaddleOCR-VL, a SOTA resource-efficient vision-language model for document parsing with 109 language support, and PP-OCRv5 multilingual recognition model with only 2M parameters.

Oct 16, 2025