v0.37.0
QwenLM/Qwen3-VLv0.37.0Sep 8, 2026by github-actions[bot]
Upgrading?
AI Summary
This update introduces higher-quality video recording, WebMCP availability reporting, and session setup inheritance for newly opened tabs.
Key Highlights
- Added higher-quality video recording (30 fps default, configurable 1-60 FPS)
- Added WebMCP availability output to navigation responses
- Added session setup inheritance for new tabs (headers, credentials, user agent, etc.)
- Fixed recording startup validation and navigation state handling
New Features
- Video recording improvements with `Page.startScreencast`
- WebMCP availability metadata for CLI and JSON clients
- Session inheritance for active session properties
Full Release Notes
### New Features - Added **higher-quality video recording**: `record start` and `record restart` now capture the current active page at 30 fps by default, support `--fps 1-60`, use `Page.startScreencast` for smoother motion, and preserve wall-clock timing. WebM and MP4 output are documented, and `doctor` reports the ffmpeg recording dependencies (#1763, #1776, #1778) - Added **WebMCP availability output** so navigation responses advertise when a page exposes allowed WebMCP tools, including availability metadata for CLI, JSON, and MCP clients (#1760) - Added **session setup inheritance for new tabs**. Tabs opened with `tab new` or through page clicks now inherit the active session's headers, credentials, user agent, locale, timezone, geolocation, offline mode, routes, color scheme, and init scripts before their first navigation (#1777) ### Bug Fixes - Fixed **recording startup validation** so missing ffmpeg, extensionless output paths, and invalid recording options fail before browser or recording state changes. Failed replacements preserve the active take, and ffmpeg errors now include useful diagnostics (#1778) - Fixed **recording navigation state** so URL navigation during recording clears stale element refs, frame scope, and page WebMCP state like normal navigation (#1776) ### Contributors - @jamesvclements - @ctate - @Railly
Related Repositories
rasbt/LLMs-from-scratch★ 107,049
Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
lyogavin/airllm★ 40,163
AirLLM 2.8T LLM inference with single 4GB GPU
NirDiamant/RAG_Techniques★ 29,958
This repository showcases various advanced techniques for Retrieval-Augmented Generation (RAG) systems. Each technique has a detailed notebook tutorial.
microsoft/OmniParser★ 25,539
A simple screen parsing tool towards pure vision based GUI agent
karpathy/nn-zero-to-hero★ 24,906
Neural Networks: Zero to Hero