openpilot Releases

Reverts a previous change regarding autobatch deletions by filter with additions to address potential issues.

May 28, 2026

Adds OCR capabilities for images and PDFs, fixes memory leaks in PDF conversion, and introduces Azure Content Understanding support.

May 26, 2026

Introduces bold template packs, improved style discovery, more reliable deck output, and agent-neutral packaging instructions for various coding agents.

May 26, 2026

A major version release introducing support for new social platforms (Virgool, Discourse), improvements to existing platforms (Substack, Calendly, Medium), and updates to Python support and ontology.

May 26, 2026

Updates AI model dependencies and prepares for the project's end by sunsetting project routes and updating documentation.

Apr 24, 2026

Focuses on infrastructure improvements including CI runner updates and fixes for cross-compilation compatibility.

Apr 17, 2026

Officially announces the shutdown of the Bloop company and the Vibe Kanban project.

Apr 10, 2026

A major feature release adding support for over 30 new social media platforms and improving the build system.

Apr 9, 2026

Apr 3, 2026

The driving model is updated to be fully trained using a learned simulator, improving longitudinal performance in Experimental mode. Power usage for comma four standby is significantly reduced. Support is added for the Kia K7 2017 and Lexus LS 2018.

Mar 17, 2026

This release implements a progressive disclosure restructure to reduce context bloat by approximately 89%, transforming the skill from 1,625 lines to a leaner architecture with on-demand loading of supporting files. The update follows best practices from Claude Code's skill documentation and OpenAI's harness engineering approach while maintaining all existing functionality.

Mar 4, 2026

This release focuses on PDF table extraction improvements and bug fixes, adding support for aligned Markdown, wide tables, and markdown in Accept headers. The update also removes Windows dependency constraints on onnxruntime and introduces new contributors to the project.

Feb 20, 2026

This release introduces a new driving model with temporal policy architecture and an on-policy training physics noise model, along with a new driver monitoring model trained on an expanded dataset including comma four data. Memory efficiency for inter-process communication has also been improved.

Dec 21, 2025

A maintenance release focused on security updates, bumping the `mammoth` library to address a specific CVE and upgrading `pdfminer.six` to resolve a security advisory.

Dec 1, 2025

This release adds comma four support to openpilot. Note that no release was shipped for comma 3X as the only changes relate to comma four compatibility.

Nov 30, 2025

This release introduces a new configuration option for tables and resolves an issue with dynamic table layouts.

Nov 21, 2025

This release features a major new driving model with significant architecture changes including a doubled parameter count for the World Model and 4x more training segments. A new driver monitoring model is also introduced, along with support for multiple new Honda and Acura vehicle models.

Oct 30, 2025

This major release introduces a completely new driving model training architecture based on the CVPR paper 'Learning to Drive from a World Model', replacing longitudinal MPC with E2E planning in Experimental Mode. Live-learned steering actuation delay is now enabled, and opt-in audio recording for dashcam video is available.

Aug 21, 2025

This release focuses on security, infrastructure improvements, and developer experience. It includes an XSS vulnerability fix, integration of Claude AI for automated code reviews, and enhanced documentation for the project.

Jul 9, 2025

This release introduces a new driving model using components from MLSIM with online learning for steering actuation delay. Support is added for multiple new vehicle models including Ford, Hyundai, Tesla, and Lexus vehicles.

Jun 19, 2025