release
Apr 16, 2026
By Teun
OpenClaw 2026.4.15 Ships with Claude Opus 4.7 and Cloud-Backed Memory
Latest release makes Opus 4.7 the default Anthropic model, adds Gemini text-to-speech, cloud-backed LanceDB memory, and a new Model Auth status card showing OAuth health at a glance.
OpenClaw 2026.4.15 is out, and the most visible change is that Claude Opus 4.7 is now the default model when Anthropic is selected as the provider. According to the release notes, aliases and CLI defaults now point to Opus 4.7, so users do not need to reconfigure their setup after upgrading.
The update also keeps bundled image understanding enabled, which matters for agents that need to inspect screenshots, documents, or other visual inputs. For people running OpenClaw as part of a larger automation flow, that means the default Anthropic path now ships with the newer model and the same multimodal behavior already built in.
⚡ New to this?
OpenClaw has updated its default Anthropic model to Claude Opus 4.7 and added a few features that make agents easier to run, including voice output, cloud-backed memory, and a status check for model authentication. A vector database, or LanceDB in this case, stores and retrieves past context for an agent so it can remember useful information across sessions. Non-specialists should care because these changes affect how reliably an AI agent can talk, remember, and keep working without manual fixes.
🦞 OpenClaw angle
If you run OpenClaw in production, upgrade in a staging environment first and verify that the Anthropic default now points to Opus 4.7 without breaking prompts or cost assumptions. If you depend on persistence across workers or containers, move your memory layer to cloud-backed LanceDB and test that retrieval still works after restarts. Also add the new Model Auth card to your regular checks so expired OAuth tokens or rate-limit pressure are caught before an automation fails mid-run.
Another addition is Gemini text to speech in the bundled Google plugin. That lets an agent return spoken output instead of only text, with selectable voices and audio output in WAV or PCM format, according to the release summary.
That makes OpenClaw more usable for voice assistants, interactive kiosks, and workflows where a human needs to hear a response rather than read it. It also lowers the amount of glue code needed to bolt speech onto an existing agent, since the audio path is now part of the packaged plugin.
Memory handling gets a practical upgrade too, with cloud backed LanceDB storage now supported alongside local files. LanceDB is a vector database, used here to store retrieval memory so an agent can recall prior context across sessions, and cloud backing makes that memory easier to share across machines or containers.
The release adds a new Model Auth status card in the dashboard as well. According to OpenClaw, the card shows OAuth health and rate limit pressure at a glance, which gives operators a quick check on whether model access is healthy before an agent fails mid task.
Taken together, 2026.4.15 is mostly about reducing manual setup and making core agent plumbing easier to monitor. The release is already available on GitHub, so the next step for users is to test the default model switch, verify voice output if they use Google, and confirm their memory backend and OAuth status look correct after upgrade.
