In late April, OpenClaw (NousResearch’s open-source AI Agent platform, ~105k GitHub Stars) released its 2026.4.24 version update. Three changes combined signal a redefinition of the Agent platform competitive landscape.
What Changed
1. DeepSeek V4 Flash as Default Model
New OpenClaw installations now default to DeepSeek V4 Flash instead of Claude Sonnet. V4 Pro is also available in the built-in model directory. This means new users can start using Agent capabilities at near-zero cost — community testing confirms V4 Flash’s cost-effectiveness makes it “fully sufficient” for daily Agent tasks.
2. Real-Time Voice Fully打通
The largest upgrade in 4.24: Talk, Voice Call, and Google Meet now support realtime voice loops with full OpenClaw Agent invocation during calls. AI no longer just “listens then speaks” — it can think and operate tools while listening.
Voice calls now include full Agent mid-call access: the Agent can query data, browse the web, or call external APIs without hanging up or switching interfaces.
3. Browser Automation: Pixel-Level Clicking
OpenClaw now supports viewport-coordinate-based clicking, enabling precise coordinate interaction on web pages. This covers scenarios where DOM element targeting falls short (Canvas, custom UI components).
4. WhatsApp Voice Message Transcription
The new version supports WhatsApp voice message transcription, expanding OpenClaw’s messaging integration capabilities.
Comparison
| Capability | Before 4.24 | After 4.24 |
|---|---|---|
| Default model | Claude Sonnet | DeepSeek V4 Flash |
| Real-time voice | Limited support | All channels (Talk/Voice Call/Meet) |
| Agent in voice calls | Not supported | Real-time mid-call access |
| Browser clicking | DOM element targeting | New viewport pixel-level clicking |
| WhatsApp voice | Not supported | Transcription supported |
Quick Start
If you already have OpenClaw installed:
# Update to latest version
openclaw update
# Check current version
openclaw --version # Should show 2026.4.24New users can install directly — DeepSeek V4 Flash is pre-configured as the default:
# Fresh install
openclaw installObservations
Setting DeepSeek V4 Flash as the default model is a notable strategy: it dramatically lowers entry costs while reducing dependency on a single model provider. However, V4 Flash’s stability in complex Agent workflows still needs more real-world testing.
The integration of real-time voice with full Agent capabilities moves “natural language dialogue to control Agent workflows” from concept to usable. Whether this becomes mainstream in Agent interaction paradigms depends on further optimization of response latency and accuracy.