openclaw v2026.6.10 release notes: what's new & changelog

openclaw v2026.6.10 delivers automatic fast mode for short conversational turns, more consistent model routing for Zai and GLM, and improved reliability for session and channel state management.

Key Changes

Automatic Fast Mode for Conversations

OpenClaw now supports fast: auto mode, which enables fast mode for short conversational turns and automatically returns to normal mode for longer runs. This system uses a default 60-second cutoff (configurable via fast_seconds) and emits channel-visible progress notices (e.g., 💨Fast: auto-off(...) and 💨Fast: auto-on) at tool-execution boundaries. This feature is integrated into the Telegram /fast command, TUI, and UI.

Enhanced Model Routing and Reasoning

Model synthesis and failover mechanisms are now more reliable and consistent with the active model catalog:

• Zai and GLM Improvements: GLM-5.2 now exposes full reasoning levels (off, low, high, max), with low/high mapping to reasoning_effort: "high" and max mapping to reasoning_effort: "max".
• Failover Correctness: Zhipu GLM overload errors (specifically the 访问量过大 phrase) are now correctly classified as overloaded, triggering the appropriate failover recovery semantics.
• Base URL Fallbacks: Synthesized GLM-5 models now fall back to the manifest baseUrl when provider configuration or template models are unavailable, preventing incorrect routing to api.openai.com.
• Dynamic Reasoning Menus: Native /think menus in Telegram, Slack, and Discord now resolve against the runtime catalog, ensuring that live-discovered reasoning models (such as those from Ollama) display all available thinking levels.

Session and Channel State Reliability

• Cross-Channel State Reset: In dmScope: "main" sessions, switching channels (e.g., moving from Slack to Telegram) now resets stale per-channel origin fields such as nativeChannelId and threadId. This prevents agents from referencing the previous channel's context when operating in a new channel.
• Cron Delivery Awareness: Scheduled cron deliveries now preserve awareness for target sessions. When a cron job delivers a message to a channel, the receiving channel's next agent turn is injected with system-event context about the delivery, allowing the agent to know what was just sent without leaking cron metadata into the user-visible message.

Security and Plugin Infrastructure

• Trusted Policy Persistence: Composed hook registries now retain trusted tool policies during hook composition. This ensures that approval-sensitive flows remain protected even when a Gateway registry is pinned and later registries are initialized.
• Onboarding Plugin Refresh: The first-time onboarding process now refreshes the provider plugin registry immediately after installing an external provider plugin. This resolves a loop where users were prompted to install a plugin but were not immediately directed to the credential/API key prompt.
• SDK Transcript API: A new combined public SDK transcript runtime API provides stable transcript identity and scoped helpers for reading, writing, publishing, and locking session transcripts.

Impact

Users will experience more responsive conversational turns through automatic fast mode and more accurate model behavior when using Zai and GLM reasoning models. System administrators and developers will benefit from more reliable session targeting across multiple channels and a more seamless onboarding experience for new provider plugins. Security is tightened for tool-call hooks through the preservation of trusted policies across composed registries.

FAQ

What's new in v2026.6.10?

This release introduces fast: auto mode for conversational turns, expanded reasoning controls for GLM-5.2, runtime-catalog-aware /think menus for live-discovered models, and fixes for cron delivery awareness and cross-channel session state.

Are there any breaking changes?

The provided source material does not mention any breaking changes.

How do I upgrade?

The source material does not provide specific upgrade instructions beyond the standard release process.