OpenClaw v2026.4.15-beta.2 delivers a focused beta update that improves default model behavior for Anthropic users, expands multimodal voice capabilities through Gemini text-to-speech, and strengthens gateway-level safety around trusted media tool passthrough. For teams using AI agents in production workflows, the release is most notable for better default Claude routing and a more capable bundled voice stack.
The release updates Anthropic model handling across the platform. Default Anthropic selections have been refreshed, opus aliases have been adjusted, Claude CLI defaults were updated, and bundled image understanding now points to Claude Opus 4.7. This means users relying on Anthropic-backed tasks should see more consistent default behavior without needing as much manual configuration.
OpenClaw also adds Gemini text-to-speech support to the bundled google plugin. The update includes provider registration, voice selection, WAV reply output, PCM telephony output, and setup documentation. This expands the platform's native voice output options and makes it easier to support both standard audio replies and telephony-oriented audio pipelines.
For AI operations teams, the Anthropic default changes reduce friction in day-to-day usage. When defaults and aliases are aligned more cleanly, agent behavior becomes easier to predict across CLI and plugin-driven workflows. The bundled move to Claude Opus 4.7 for image understanding is especially relevant for users who want stronger multimodal performance out of the box.
The Gemini text-to-speech addition is significant because it broadens OpenClaw's built-in support for voice experiences without requiring separate external glue code. WAV and PCM telephony output support suggest practical use cases in customer support, voice bots, and internal automation systems that need reliable audio generation in different formats.
On the security and reliability side, the gateway fix tightens how trusted local MEDIA: passthrough is validated. By anchoring trust to the exact raw names of this run's registered built-in tools and rejecting normalized client tool definitions, the update reduces the risk of ambiguity or spoofing in tool-result handling. That is a small changelog line, but it matters for any environment where tool execution boundaries need to stay strict.
Official Source: https://github.com/openclaw/openclaw/releases/tag/v2026.4.15-beta.2