Should you update OpenClaw v2026.11.3? — Update with care
Verdict: Update with care (high confidence), assessed 2026-11-15.
Evidence: 19 credible blocking issues confirmed for this version, none widespread.
Best version to run today: v2026.11.2 — 7 days in the field, 1 credible known issue, none widespread; last verdict Update with care. If you are on a newer version that works for you, stay put. The latest, v2026.11.3, came out 5 days ago — too new to call settled.
Update with care: multiple confirmed critical bugs across macOS, Windows, Linux and the web UI — none widespread, but back up config and state first.
Why this verdict
v2026.11.3 ships meaningful feature and hardening work (private task results, reliable child-task handoffs, bounded authentication, verified workspace connections per the changelog) but carries a heavy load of credible, open blockers: the evidence gate counts 19 confirmed for this version, none widespread. Six of the tier-1 blockers are rated critical: #196820 (a confirmed regression) locks the Control UI into a 100% CPU Element.getAnimations() loop reproduced in two independent engines (macOS WebKit WebContent and Edge 152); #197806 is a hard SIGTRAP crash of the macOS app on first Talk Mode use, before permissions are even granted; #198191 silently loses long Control UI dictations because stop only waits 8 s for the final transcript; #196958 permanently destroys an OpenAI OAuth sign-in when a token refresh runs while offline; #198160 can permanently freeze config writes when a re-indented config trips the last-known-good size check, with no CLI reset path; and #199614 starves gateway RPC at ~100% CPU for hours during the transcript-facts backfill on large SQLite databases, restarting from zero after every gateway restart. High-severity tier-2 issues add to the load: #198475 (Template Add fails on a completed agent-deletion tombstone), #198050 (a delivered silent cron run is falsely marked failed and re-reports), #197337 (gateway shutdown hangs on SIGTERM with an isolated recurring cron job), #196544 (silent 8192-token output cap truncating tool calls), plus update-path failures #196987 and #197399 and the doctor/update livelock #198218.
Known issues (60+)
- #196820 (critical) [Bug]: Control UI locks up at ~100% CPU in a `Element.getAnimations()` loop (macOS app WebContent and Edge)
- #197806 (critical) macOS app crashes (SIGTRAP) when starting Talk Mode — Swift concurrency assertion in PermissionManager.ensureSpeechRecognition
- #198191 (critical) Control UI dictation: long recordings are silently lost because stop waits only 8 s for the final transcript
- #196958 (critical) OpenAI ChatGPT sign-in is lost when a token refresh runs while the network is down
- #198160 (critical) [Bug]: last-known-good size-drop check counts whitespace; a re-indented config permanently blocks all later writes (no way to reset the baseline)
- #199614 (critical) perf(sessions): transcript facts backfill restarts from zero and starves gateway RPC on large databases
- #198386 (high) [Bug]: Regression in 2026.11.3 — completion replies reject media again
- #197531 (high) macOS TTS: reported truncated CAF from segmented MP3 despite successful conversion
ClawStat.us gives an independent, evidence-based answer to whether you should update OpenClaw — for every release. Is the latest OpenClaw update safe, or should you wait for the next one? Each verdict is built from post-release bug reports and a two-model review, refreshed every few hours.
Machine-readable: JSON API · llms.txt · full markdown
Independent, unofficial project — not affiliated with or endorsed by OpenClaw or its maintainers. Verdicts are generated automatically and may be wrong; confirm against the linked issues before updating. Data is drawn from the public github.com/openclaw/openclaw issue tracker and releases via the GitHub API.