OpenClaw v2026.5.16-beta.1 Release Summary

OpenClaw v2026.5.16-beta.1 brings a substantial wave of stability improvements, provider-specific fixes, and quality-of-life enhancements. This release focuses heavily on refining the agent runtime, improving the reliability of external channel deliveries, and expanding accessibility through CLI localization.

Key Changes

CLI & Onboarding

• Localization Support: The CLI setup and onboarding wizard now support English, Simplified Chinese, and Traditional Chinese. Users can select their preferred locale using the OPENCLAW_LOCALE environment variable or by relying on system locale settings.

Agents & Skills

• Skill Hydration Caching: To reduce redundant rebuilds, hydrated resolvedSkills are now cached across warm gateway turns. This cache is keyed by a redacted configuration fingerprint, ensuring that config-gated skill boundaries are respected without leaking secrets.
• Provider Fixes (Xiaomi & OpenRouter):
  • Xiaomi MiMo: Fixed a critical issue where mimo-v2-pro and mimo-v2-omni models appeared blank because they returned final answers in the reasoning_content field. Additionally, multi-turn tool calls now correctly include reasoning_content to prevent 400 Param Incorrect errors.
  • OpenRouter: Fixed a regression where DeepSeek V4 models returned HTTP 500 errors on turns following tool calls due to empty-string reasoning_content placeholders being replayed.
• OpenAI Compatibility: The system now honors compat.supportsUsageInStreaming for OpenAI-compatible providers, ensuring that stream_options.include_usage is only sent to supported endpoints (e.g., Volcengine).

Channel & Integration Updates

• Telegram: Introduced opt-in messages.groupChat.ambientTurns: "room_event" handling. This allows always-on ambient chatter to run as quiet room context, speaking visibly only when the agent explicitly uses the message tool.
• Slack:
  • Outbound link unfurling is now disabled by default to reduce noise in channels.
  • Fixed a bug where tool failure warnings could overwrite finalized assistant drafts during streaming.
• LINE: Resolved a silent push failure where long-running tasks failed because recipient IDs were being lowercased during session-key recovery. The system now rejects malformed IDs early to prevent silent retry loops.
• Twitch: Fixed an auto-restart loop where the channel supervisor interpreted a successful startup as a clean exit.

System & Infrastructure

• Memory Search Optimization: Removed chokidar write-stability polling for memory and QMD watchers. This eliminates a significant file descriptor (FD) leak that previously caused spawn EBADF errors when watching large Markdown trees.
• Security Hardening: Queued system events now render as plain System: lines. Authority (owner vs. non-owner) is now handled via structured metadata (forceSenderIsOwnerFalse) rather than model-visible text, preventing trust-marker spoofing.
• Gateway Lifecycle: Raised default timeout budgets for lifecycle hooks: 5 seconds for gateway:shutdown and 10 seconds for gateway:pre-restart.

Impact

Fixed Issues

• #60261: Xiaomi MiMo models outputting to reasoning_content instead of content.
• #81419: Xiaomi MiMo reasoning content passthrough missing for multi-turn tool calls.
• #82150: DeepSeek V4 via OpenRouter returning 500 errors after tool calls.
• #81628: LINE delivery-recovery sending lowercased recipients.
• #60071: Twitch plugin auto-restart loop.
• #77327 / #78224: Gateway FD leaks caused by memory search watchers.
• #82132: Node exec approval replay failures for WebChat.
• #81903: Slack live drafts replaced by tool failure status.

Performance & Stability

Users with large knowledge bases (thousands of Markdown files) will see significantly improved gateway stability due to the FD leak fix. Agent response times for warm turns are improved via the new skill hydration cache. Providers using Xiaomi MiMo or DeepSeek V4 via OpenRouter will experience a return to stable multi-turn tool usage.

Upgrade Guide

Configuration Changes

• Slack Link Previews: If you rely on link unfurling in Slack, you must now explicitly enable it in your config:

  channels:
    slack:
      unfurlLinks: true
  

• CLI Locale: To change the onboarding wizard language, set the OPENCLAW_LOCALE environment variable (e.g., OPENCLAW_LOCALE=zh-CN).
• Silent Replies: Operators can now use agents.defaults.silentReply.group: "disallow" to ensure that runner failures in group chats are visible rather than silently dropped.