OpenClaw 2026.5.2 adds Grok 4.3 default chat and plugin fixes
OpenClaw 2026.5.2 shipped Grok 4.3 as the default xAI chat model and a broad plumbing pass for plugins, session paths, messaging bridges, and voice features. The release matters because it trims startup stalls and cleans up common integration edges in self-hosted agent setups.

TL;DR
- openclaw's release thread framed OpenClaw 2026.5.2 as a reliability pass, but the most concrete default change is that openclaw's xAI note moves Grok 4.3 into the bundled catalog and makes it the default xAI chat model.
- According to openclaw's plugin summary and the Manage plugins docs, the plugin stack got an npm-first install path, beta fallback handling, doctor repair, dependency-state reporting, and less runtime loading on hot paths.
- openclaw's hot-path breakdown lines up with the GitHub release, which says startup, session listing, prompt prep, plugin loading, and large-config handling all got trimmed.
- Messaging fixes are unusually broad for a point release: openclaw's Telegram post covers forum-topic session routing and chunked sends, while openclaw's WhatsApp post says newsletter targets now preserve channel metadata instead of routing like DMs.
- The media layer also got cleanup across openclaw's voice thread and the web search docs, including TTS and realtime error surfacing, voice-call routing, and sturdier search-provider handling across Brave, Exa, Firecrawl, Gemini, Kimi, and others.
You can read the full release notes, check the xAI provider page, skim the plugin workflow docs, and dig into the Codex harness docs. One useful bit buried outside the headline: a recent blocker report described 30 to 60 second chat-turn overhead from per-turn plugin runtime work, which makes this release's repeated focus on hot paths look less cosmetic.
Grok 4.3
OpenClaw slipped in a nice default change here. openclaw's xAI note says Grok 4.3 is now bundled and becomes the default xAI chat model, and the release page says the same in the changelog.
The xAI provider docs still show xai/grok-4 as the config example, but they also document a broad bundled catalog: Grok 3, Grok 4, Grok 4 Fast, Grok 4.1 Fast, Grok 4.20 Beta, and Grok Code. The same xAI key can also power web_search, x_search, and remote code_execution, so the default-model switch touches more than chat.
Plugins and hot paths
The meat of 2026.5.2 is plumbing. openclaw's plugin summary and openclaw's hot-path breakdown both describe a pass over install flows, dependency repair, descriptor caching, and broad plugin loading.
The practical changes break down cleanly:
- Plugin lifecycle: npm-first installs, beta fallback, stale-install handling, doctor repair, and ClawHub metadata, per openclaw's plugin summary and the plugin docs.
- Runtime visibility:
openclaw plugins list --jsonnow exposesdependencyStatus, but the docs note that inventory does not prove a running Gateway actually imported the runtime. - Latency work: the release notes call out leaner startup, session listing, task maintenance, prompt prep, plugin loading, tool descriptor planning, and filesystem guards.
- Why now: a blocker issue described 75 to 80 seconds of per-turn plugin overhead in late April, and another issue traced heavy repeated filesystem work during gateway spikes.
Messaging bridges
The channel work is spread across half a dozen bridges, but the useful detail is that OpenClaw fixed session identity bugs, not just cosmetic UI glitches.
A few concrete examples:
- Telegram forum topics: native commands now bind to the right session files, group command menus cover more scopes, timeouts were relaxed, and long sends get chunked, per openclaw's Telegram post and the Telegram docs.
- WhatsApp newsletters:
@ newsletteroutbound sends now keep channel session metadata instead of falling back to DM-like routing, according to openclaw's WhatsApp post. - Slack and Discord: openclaw's Slack post lists DM routing, missing-scope errors, multi-workspace sends, uploads, and exact reads, while openclaw's Discord post lists restart-safe interactions, media handling, thread fixes, and rate-limit cleanup.
Codex and voice paths
One more useful subsystem got its own cleanup pass: Codex runtime behavior and voice observability.
The Codex harness docs make the architecture explicit. Codex owns native agent-session operations like thread resume, compaction, and app-server execution, while OpenClaw keeps channels, approvals, media delivery, and transcript mirroring. In 2026.5.2, openclaw's Codex post says OpenClaw tightened that path with native-first app-server tools, visible replies through message, clearer ChatGPT and Codex setup docs, and prompt snapshots to catch drift.
On the voice side, openclaw's voice thread says explicit TTS blocks now work reliably, realtime errors surface in Talk, local audio survives Telegram delivery, and voice calls have better routing and state handling. The Google Meet plugin docs add another concrete piece: Twilio join phases, DTMF, listen tests, and caption-health checks are now easier to debug.