Faster & More Consistent
OpenClaw 2026.6.10 focuses on snappier conversational performance through automatic fast-mode transitions and significant hardening of model routing and session state.
Automatic Fast Mode for Talks
The new automatic mode for voice and talk sessions optimizes response times based on the turn length.
- Conversational Snappiness: Fast mode is enabled automatically for short turns, providing near-instant feedback for casual greetings or simple questions.
- Dynamic Fallback: The system automatically returns to normal mode for longer, more complex runs, maintaining bounded delivery behavior.
- Persistent State: Fast-mode state now survives retries, fallback transitions, and progress events.
Reliable Model Routing
We've improved how the Gateway handles model discovery and failover paths.
- Zai & GLM Consistency: Model synthesis for Zai and failover paths for GLM now follow the active model catalog more accurately.
- Reasoning Controls: Native reasoning-level selection is resolved via the runtime catalog for live-discovered models.
- Failover Hardening: Overload classification for Zhipu GLM has been refined to ensure smoother failover transitions.
Safer Session & Channel State
Hardening the boundaries between different messaging channels and automation triggers.
- Origin Field Resets: Stale per-channel origin fields are now correctly reset during channel switches.
- Cron Awareness: Delivery awareness for scheduled tasks stays strictly attached to the target session context.
- Plugin Metadata: The provider plugin registry is now refreshed immediately after new plugins are installed during setup.
Hook Policy Persistence
Ensuring security policies remain active across complex workflow compositions.
- Composition Safety: Composed hook registries now retain the trusted tool policies required by approval-sensitive flows.
- Policy Correctness: Fallback cutoffs and reset notices are now bounded, ensuring repeated progress events remain visible.
Technical Notes
- SDK Transcript Identity: New API for managing transcript identity in the OpenClaw SDK.
- Copilot Parity: Completed harness lifecycle parity for GitHub Copilot agent runtimes.
- Doctor Health: Improved ordering of core health checks in the doctor command.
Upgrade Guide
Who should upgrade?
- Users utilizing voice or Talk mode for frequent interactions
- Operators running multi-channel gateways with frequent session switches
- Developers building complex hook-based automations
How to Upgrade
Update your installation via the terminal:
openclaw update
FAQ
What is the new 'Fast Talks' automatic mode?
It allows OpenClaw to automatically enable fast mode for short conversational turns and then return to normal mode for longer tasks, providing snappier responses for casual chat.
How has model routing been improved?
Model routing for Zai synthesis, GLM failover, and native reasoning levels is now more consistent and follows the active model catalog more accurately.
What does 'safer session state' mean?
OpenClaw now resets stale origin fields when switching channels and ensures cron delivery awareness stays correctly attached to the target session.
What's the fix for hook composition?
Composed hook registries now correctly keep the trusted tool policies required for approval-sensitive flows, preventing unauthorized tool blocks.
Need help from people who already use this stuff?
Join the OpenClaw Community
Get help with your setup, share your agent workflows, and stay up to date with the latest releases in our profit-lab community.