A 5ms response time is just outrageous; a 4100x improvement makes it feel like the gateway is turned into local cache.

OpenClaw released v2026.5.22, after gateway warm-up authentication, the model list response time dropped from about 20 seconds to 5 milliseconds, bypassing cumbersome discovery processes in the hot path, improving by approximately 4100 times. Added a standalone external plugin meeting-notes, supporting Discord voice input, accessible via command line to record audio or import text. To reduce context overhead and prevent information leakage in multi-Agent deployment, child Agents only transmit the context of AGENTS.md and TOOLS.md during startup, with personal settings, memories, and keys isolated. Upgraded underlying dependencies to protobufjs 8.4.0, integrated xAI with Grok authorization files for web search, used native interfaces for OpenAI image generation, resolving backend authentication 401 issues.