OpenClaw v2026.5.16-beta.2 landed on May 16, 2026 at 6:29 AM America/Chicago time, which was May 16, 2026 at 11:29 UTC. The shortest version is this: if you use xAI, cron, or any setup that has ever been knocked sideways by weird saved state, this beta is a pretty useful cleanup pass.

The shiny headline is xAI OAuth for SuperGrok subscribers. The practical headline is broader. OpenClaw also gets a real cron run --wait mode for manual runs, quieter ambient behavior for Telegram group rooms, and a long list of hardening fixes around malformed persistence, spoofed media payloads, and provider responses that used to fail in dumb ways.

xAI login gets much less annoying

If you have been curious about xAI models but did not want one more API key ceremony in your life, this release helps. SuperGrok subscribers can now authenticate xAI providers with OAuth, which means xai/* models and the related media/tool providers can work without an XAI_API_KEY.

That matters because it removes one of the more brittle setup steps. Fewer copied secrets. Fewer env-var mistakes. Less "why is this provider missing again" energy. I like changes like this because they save time twice: once during setup and again the next time you move machines or rebuild a node.

Cron finally gets a wait mode that fits real automation

The other big operator win is openclaw cron run --wait. You can now kick off a manual cron run and wait on that exact run id with timeout and poll controls instead of building your own awkward wrapper around loose queue state.

If you test scheduled jobs, run release checks, or glue OpenClaw into another automation layer, this is one of those changes that sounds small until you use it once. Then you wonder why it was ever missing.

The setup story is better, and group rooms get quieter

This beta also localizes the setup wizard and bundled channel setup flows for English, Simplified Chinese, and Traditional Chinese. That is not just a nice-to-have. It lowers the odds that the first run goes sideways for somebody who was already translating config ideas in their head.

On the group-chat side, Telegram now has an opt-in messages.groupChat.ambientTurns: "room_event" mode. That lets always-on ambient chatter stay as quiet room context and only speak visibly through the message tool. In plain English: the assistant can keep listening without acting like it owns the room.

A lot of ugly failure modes got boxed in

The longest section of the release notes is the least glamorous and maybe the most valuable. OpenClaw now strips or skips malformed persisted rows across sessions, cron jobs, auth profiles, task metadata, device auth, and pairing state instead of happily restoring garbage and breaking later. Media staging got stricter about files pretending to be images. Provider integrations got better at rejecting malformed "successful" responses instead of timing out or returning nonsense.

That is a lot of plumbing work. It is also the kind of work that keeps a normal operator from losing an hour to a corrupted row or a mislabeled upload. You usually do not celebrate these fixes on day one. You celebrate them the first time nothing weird happens.

What I would test right after updating

1. If you use xAI, log in the new way. Confirm your SuperGrok-backed xAI provider works without an XAI_API_KEY, then run one model call and one media or tool workflow if that matters to your setup.
2. Run one manual cron job with --wait. Make sure the exact-run blocking, timeout behavior, and exit status match what your wrapper scripts expect.
3. If you run Telegram group rooms, test ambient behavior deliberately. Turn on room_event and confirm the assistant keeps context quietly instead of spraying visible chatter into the room.
4. Restart once and watch for stale-state weirdness. This release specifically fixes restart, replay, and persistence edges, so it is worth validating the boring paths.
5. Try one sketchy file input if your workflows depend on media. The media sniffing hardening is good, but if you have custom pipelines that relabel files, better to learn that now than during a real run.

Should you install this beta now?

Yes, if xAI is part of your stack, you rely on cron-driven checks, or you have ever been burned by corrupt session state, malformed provider replies, or restart replay weirdness. This beta fixes real operator pain.

You can wait a moment if your setup is calm and none of those areas touch you. It is still a beta. I would update with intent, run the short retest list above, and move on.

Overall, this one feels less like a flashy feature drop and more like a release that removes friction from daily use. I am into that.

---

Full changelog: View v2026.5.16-beta.2 on GitHub

Generated release page: Read the structured release notes

- Fred