OpenClaw v2026.5.24-beta.1 was published on May 24, 2026 at 9:42 AM America/Chicago time, which was May 24, 2026 at 14:42 UTC. This was the first 5.24 beta, and it is the kind of release I would actually bother testing if your setup feels heavy at startup or clumsy during live voice work.

The headline is easy to explain. OpenClaw gets much faster at model browsing, voice callers can manage a run while it is still active, and the project finally has a real meeting-notes path instead of a vague promise. There is a lot more in the changelog, but those are the parts normal operators should notice.

What users can now do

Browse models without waiting twenty seconds for the list to wake up. The release notes put a number on it: the hot path for /models and similar calls drops from about 20 seconds to about 5 milliseconds after startup prewarm. That is a blunt improvement. If model browsing has been part of your daily annoyance budget, this beta goes straight at it.

Manage a live consult from WebUI or Discord voice while it is still running. You can now ask for active run status, cancel the run, steer it, or queue follow-up work before the current consult finishes. That matters more than it sounds. In voice, the worst feeling is knowing the run drifted and having no clean way to intervene.

Start using the first real meeting-notes workflow. This beta adds a source-only external meeting-notes plugin, a source-provider contract, manual transcript imports, read-only openclaw meeting-notes CLI access, and Discord voice as the first live source. It is early, but it is no longer imaginary.

Choose how aggressive image compression should be. The image tool now understands a model-aware compression path, plus an agents.defaults.imageQuality preference for token-efficient, balanced, or high-detail handling. That gives heavy screenshot users a real knob instead of a fixed compromise.

Save provider auth under a specific profile id. openclaw models auth login --profile-id can now keep a single returned provider auth profile under the name you asked for, which is cleaner if you juggle multiple logins.

What got safer or less annoying

Gateway startup and hot paths got a lot of low-level cleanup. The release caches install records, channel catalog reads, plugin metadata snapshots, and SDK alias maps, while lazy-loading handler trees and startup-idle work. You do not need to care about every detail to care about the outcome. Gateway should spend less time rereading the world.

Delegation also got tighter. Subagents now start with a much smaller default bootstrap context, which keeps persona, memory, heartbeat, and setup files out of the child unless you ask for them. That lowers the odds of a delegated task carrying extra baggage into a job that should have been simple.

Packaging and install behavior look saner too. The npm tarball drops documentation assets that were only taking up space, and the root package plus OpenClaw-owned npm plugins now ship with generated shrinkwrap so installs use a locked dependency graph.

What I would test right after updating

1. Open your usual model picker flow. Run /models or the UI equivalent after a fresh startup and see whether it feels instant on your machine.
2. Start one live consult and interrupt it on purpose. Ask for status, steer it, cancel it, or queue follow-up work before it finishes.
3. If you use Discord voice, try the meeting-notes path on a real conversation. Confirm capture, transcript routing, and read-only CLI access all line up.
4. Send one image-heavy request that matters. Check whether the new image-quality preference gives you the tradeoff you actually want.
5. Delegate one subagent task that used to over-inherit context. Make sure the child still has what it needs, and only what it needs.

Should you install this beta?

Yes, if your pain is in models, voice control, or startup drag. There is real operator value here, especially for people running large provider sets, Discord voice consults, or early meeting-notes experiments.

If your setup is tiny and steady, you can wait for a later cut. But v2026.5.24-beta.1 looks like a meaningful beta, not filler.

Full changelog: View v2026.5.24-beta.1 on GitHub

Generated release page: Read the structured release notes

- Fred