OpenClaw new version reduces model list latency by 4000 times and supports Discord voice conference recording According to Beating monitoring, the open-source AI assistant project OpenClaw has released version 2026.5.22, which brings important updates in gateway performance and external plugin support. The new version significantly reduces the response delay of the model list acquisition interface from about 20 seconds to 5 milliseconds by pre heating the authentication status of the thermal service provider during gateway startup, achieving a performance leap of nearly 4100 times. The state preheating mechanism allows model listing requests in the hot path to bypass the cumbersome plugin and external command line discovery process directly. In terms of plugin ecology, the new version has launched for the first time an external plugin for meeting nodes that is independent of the core npm package, and supports Discord voice channel as the first real-time input source. Users can access recordings through the read-only command-line tool openclaw meeting nodes, or manually import recordings as text. In order to reduce the context overhead during multi-agent deployment and prevent sensitive information leakage, the new version has also restructured the startup context mechanism of sub agents. By default, only AGENTS.md and TOOLS.md are transmitted to delegated sub agents, completely isolating personal settings, user memory, and identity key files. In terms of underlying dependencies and ecological integration, the project will upgrade the protobufjs dependency library to version 8.4.0 to fix security vulnerabilities. For xAI integration, the new version supports reusing Grok's OAuth authorization files for web search. At the same time, the new version re routes OpenAI image generation to the native image interface, completely solving the historical defect of 401 error caused by failed authentication of legitimate API keys in the backend of Codex streaming. [Original link]