OpenClaw Adds Kimi K2.6 Support and Provider Aware Reasoning Controls

OpenClawOpenClaw

· Updated

OpenClaw v2026.4.20 introduces native support for Moonshot's Kimi K2.6 and a unified reasoning logic that normalizes internal thinking tokens across different model providers. This update stabilizes agentic flows on messaging platforms like iMessage while adding tiered pricing for more accurate token cost estimates.

OpenClaw, an open-source self-hosted AI assistant, released version 2026.4.20 with native support for Moonshot's kimi-k2.6, mirroring the rapid adoption of the long-horizon model across inference providers. This model now handles bundled web search and media-understanding tasks. The update also introduces provider-aware reasoning that normalizes /think settings across different model families.
Model support
Kimi K2.6, Kimi K2.5
Messaging platforms
BlueBubbles, Mattermost, Telegram, Slack, Matrix
Reasoning control
Provider-aware /think logic
Security
Device pairing restrictions, plugin dependency isolation
OS compatibility
macOS 26 Tahoe send timeout fixes
Cost tracking
Tiered model pricing for cost estimates

As reasoning models become standard, orchestrators must handle non-standardized thinking tokens. This release follows recent reliability updates for GPT-5.4 by ensuring reasoning efforts are correctly routed or stripped when unsupported. It addresses connectivity issues that often cause autonomous agents to stall when communicating through messaging bridges.

You can now deploy kimi-k2.6 for search-heavy workflows or use the improved BlueBubbles integration for reliable iMessage interactions. The update is available on GitHub, featuring security hardening and isolated plugin dependencies, a foundation that extends to the automated repair tools in v2026.4.21. The system also supports tiered pricing for precise token-usage cost estimates.

View the full update on github.com
OpenClaw🦞
OpenClaw🦞
@openclaw
X

OpenClaw 2026.4.20 🦞 🧠 Kimi K2.6 support + provider-aware /think 💬 BlueBubbles iMessage sends + tapbacks fixed ⏰ Cron state/delivery cleanup 🔐 Gateway pairing + plugin startup hardening Less haunted. More useful. https://t.co/5KvQJkP0j3

36retweets353likes
View on X

Still wondering? A few quick answers below.

OpenClaw now natively supports Moonshot's Kimi K2.6 model, making it the default choice for bundled web search and media-understanding tasks. Users can still access Kimi K2.5 for compatibility. This integration allows the assistant to use Kimi's specific capabilities for browsing and analyzing media while managing internal reasoning tokens through the new provider-aware logic.

The new provider-aware reasoning logic normalizes how OpenClaw handles internal thinking tokens across different AI providers like OpenAI, Anthropic, and Moonshot. It ensures that settings like turning reasoning off or setting it to maximum map correctly to each model's specific API requirements, preventing errors when switching between models that have different reasoning effort levels or configurations.

This release addresses several stability issues for BlueBubbles, which bridges OpenClaw to iMessage. It raises the outbound message timeout to 30 seconds to accommodate macOS 26 Tahoe setups where sends may stall. It also introduces group-specific system prompts, allowing users to configure unique behavioral instructions, such as specific tapback conventions or reply styles, for different group chats.

Yes, the update introduces support for tiered model pricing within cached catalogs and configured models. This allows the system to provide more accurate cost estimates for token usage in reports, specifically including bundled estimates for Moonshot's Kimi K2.6 and K2.5. It also fixes a bug where repeated persistence paths would incorrectly compound the estimated run costs.

Security hardening in this version includes new restrictions on device pairing to prevent non-admin sessions from seeing other devices or approving unauthorized requests. It also isolates plugin dependencies into their own runtime directories to prevent conflicts and blocks untrusted workspace environment variables from overriding core settings. Additionally, the setup process now relaunches in a fresh process to protect gateway secrets.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from OpenClaw →

Share this update