OpenClaw v2026.2.21 expands its model ecosystem with Gemini 3.1, enters the Chinese AI provider market, and delivers Discord voice channel support.
Gemini 3.1 Support
Google's latest Gemini 3.1 Pro model is now available via google/gemini-3.1-pro-preview. Users can switch to Gemini 3.1 through OpenClaw's model configuration or through OpenClaw Launch's visual configurator.
Volcano Engine & BytePlus Providers
Two new providers bring OpenClaw into the Chinese AI ecosystem. Volcano Engine (Doubao), ByteDance's cloud AI platform, and BytePlus, its international variant, are now supported with coding-specific model variants. This follows Baidu's recent integration and signals growing OpenClaw adoption in China.
Discord Voice Channels
Discord bots powered by OpenClaw can now join voice channels with /vc commands and auto-join configuration. Subagents operate with thread-bound sessions and continuation routing for complex voice interactions.
Per-Channel Model Overrides
A new channels.modelByChannel configuration lets users assign different AI models to different messaging channels. For example, you could use a fast model for Telegram quick replies and a more capable model for Discord coding help.
Other Improvements
- Security — Owner-ID obfuscation now uses dedicated HMAC secrets, SHA-256 replaces SHA-1 for lock/tool-call IDs, and sandbox browser isolation adds VNC password authentication.
- iOS — Chat UI improvements including keyboard dismissal and metadata stripping.
- Watch — Actionable approval controls and quick-reply actions on Apple Watch.
- Streaming — Simplified streaming with configurable preview modes for Telegram and Discord.