OpenClaw v2026.4.22 dropped on April 22 and it's one of the meatier releases this cycle. Three major threads: new AI provider support (xAI Grok across text, image, and voice), a wave of stability fixes targeting message duplication and session interruptions, and CLI improvements that make model management significantly less painful.
Here's what changed and what it means for running OpenClaw in production.
1. Full xAI Integration: Grok for Image, Voice, and Text
xAI (Elon Musk's AI lab) is now a first-class provider in OpenClaw — and it's not just chat. The integration covers three capability areas:
Image Generation
grok-imagine-image and grok-imagine-image-pro, including reference-image editing. Generate or modify images directly in conversation.
Text-to-Speech
6 live voices. Output formats: MP3, WAV, PCM, G.711. Alternative to ElevenLabs and Deepgram for voice synthesis.
Speech-to-Text
grok-stt for batch audio transcription + real-time transcription for voice call streaming. Joins Deepgram and ElevenLabs STT options.
To use xAI, add your API key via openclaw configure or the config file, then select a grok model: openclaw --model xai/grok-3. Image generation and voice features activate automatically when you call them within a conversation.
Why this matters: xAI adds a third competitive option for image generation (alongside OpenAI's gpt-image-2 and Google's Imagen). For voice workflows, grok-stt real-time STT directly competes with Deepgram and ElevenLabs without requiring separate API accounts. One less integration to manage.
2. Real-Time STT Strengthened Across All Providers
Beyond xAI, real-time speech-to-text got cross-provider improvements in this release. OpenClaw's voice call streaming — which previously required ElevenLabs or Deepgram — now supports xAI as a third streaming STT option.
Additionally, the voice call pipeline received latency improvements. If you're running OpenClaw as a voice assistant (connected to macOS/iOS voice, or Discord voice channels), this release should noticeably improve real-time responsiveness.
3. Stability Fixes: Duplicate Messages, Session Interruptions, Config Loss
This is the section most production users will care about most. A set of bugs that caused unreliable behavior in long-running OpenClaw instances were addressed:
- Fixed Duplicate messages — Messages being sent twice in certain WhatsApp and Discord flows. Root cause: retry logic firing on delivery confirmations that were misread as failures.
- Fixed Session interruptions — Long-running agent sessions dropping mid-task without clean recovery. Particularly affected heartbeat-driven agents running overnight.
- Fixed Configuration loss on restart — Config merging logic bug caused certain plugin settings to revert to defaults after gateway restart. Now uses proper deep merge.
- Improved WhatsApp automation — Native reply quoting improved; per-group system prompts now persist correctly across sessions; message delivery confirmation more reliable.
- Fixed Plugin discovery hang — Compatibility issue with lossless-claw and similar plugins causing gateway startup to hang. Now resolved upstream.
4. Model Management CLI Improvements
Several quality-of-life improvements to the models CLI that make day-to-day model management cleaner:
models list --all— Now shows bundled provider-owned static catalog rows (including Kimi K2.6 rows for Moonshot, OpenRouter, and Vercel AI Gateway) before auth is configured. Useful for planning model strategy before committing API keys.models list --provider— Now validates that you're passing provider IDs, not display labels. Clearer error message if you get it wrong./models addfrom chat — Add new models live without restarting the gateway. Previously required a config edit + restart.- Onboarding TUI speed — The configure wizard reaches its first prompt faster when the gateway is unavailable. Removes a common frustration during fresh installs.
5. WhatsApp: Native Reply Quoting + Per-Group System Prompts
Two significant WhatsApp UX improvements worth calling out separately:
Native reply quoting — When OpenClaw responds to a WhatsApp message, it now properly uses WhatsApp's native quote-reply format. The context bubble appears just like it would from a human sender. Makes multi-turn conversations in WhatsApp groups feel natural.
Per-group system prompts — You can now configure different system prompts (personality/behavior) for different WhatsApp groups. Useful if you have OpenClaw in a work group (professional tone) and a personal group (casual). Set per-group overrides in config under channels.whatsapp.groups.
How to Update
Update to v2026.4.22
Run in your terminal:
npm update -g openclaw && openclaw gateway restart
Verify: openclaw --version should show 2026.4.22 or later. Allow 30–60 seconds for gateway restart to complete.
Security reminder: Staying current on OpenClaw releases is a security requirement, not just a feature upgrade. CVE-2026-41352 and CVE-2026-41349 (both CVSS 8.8) were patched in recent releases. If you're running anything older than 2026.3.31, update immediately.
Bottom Line
v2026.4.22 is a solid maintenance release with a genuinely useful capability addition (xAI Grok). The stability fixes are the headline for anyone running production OpenClaw deployments — duplicate messages and session drops are exactly the class of bug that makes agents unreliable in long-running overnight tasks. If you've been deferring an update, this one is worth prioritizing.
Running an Outdated OpenClaw Instance?
ClawReady's $49 audit covers version status, security config, gateway hardening, and memory architecture review. We'll tell you exactly what needs fixing — and fix it if you want us to.
Book a Security Audit →