Ai

OpenClaw v2026.4.15-beta.2 sharpens Anthropic defaults and adds Gemini speech output

OpenClaw v2026.4.15-beta.2 sharpens Anthropic defaults and adds Gemini speech output

OpenClaw v2026.4.15-beta.2 is a focused beta release that improves default model behavior, expands voice capabilities, and tightens gateway security around tool result handling. The headline changes are a move to Claude Opus 4.7 for bundled Anthropic defaults, the addition of Gemini-powered text-to-speech in the Google plugin, and a fix that strengthens how trusted local MEDIA tool outputs are validated.

What Changed

The Anthropic model layer has been refreshed so default Anthropic selections, opus aliases, Claude CLI defaults, and bundled image understanding now point to Claude Opus 4.7. This matters because it changes the baseline behavior for users relying on default Anthropic routes, reducing ambiguity around which Claude tier is being invoked and aligning multimodal understanding with the newest bundled Opus default.

The bundled google plugin now supports Gemini text-to-speech. According to the changelog, that includes provider registration, voice selection, WAV reply output, PCM telephony output, and related setup and documentation guidance. In practical terms, OpenClaw users can now generate spoken responses through Gemini in more than one output format, which makes the feature relevant for both standard voice playback and phone-oriented audio pipelines.

On the gateway side, this beta also fixes how trusted local MEDIA: passthrough is anchored for tool results. The validation now keys off the exact raw name of the currently registered built-in tools and rejects client tool definitions whose names normalize in conflicting ways. That is a meaningful hardening step because MEDIA passthrough sits close to how files and generated outputs are delivered, so stricter validation reduces the risk of unsafe or ambiguous tool-name handling.

Why It Matters

This version is less about broad platform expansion and more about improving core defaults and platform reliability. The Anthropic default shift to Claude Opus 4.7 should make model selection more predictable for teams using OpenClaw without extensive per-tool overrides.

The Gemini TTS addition is the most visible new capability in the release. It extends the built-in Google integration beyond text and model access into audio generation, which could make OpenClaw more useful for assistants, phone workflows, and voice-enabled automations.

The MEDIA passthrough fix is equally important from an operational perspective. Even though it is framed as a bug fix, it improves trust boundaries in gateway tool handling, which is exactly the kind of low-level safeguard that matters in agent platforms where tools, attachments, and generated media are closely linked.

Official Source: https://github.com/openclaw/openclaw/releases/tag/v2026.4.15-beta.2