Should you update OpenClaw v2026.10.5? — Safe to update

Verdict: Safe to update (medium confidence), assessed 2026-10-22.

Evidence: No credible blocking issue is confirmed for this version.

Best version to run today: v2026.10.3 — 9 days in the field, 2 credible known issues, none widespread; last verdict Update with care. If you are on a newer version that works for you, stay put. The latest, v2026.10.5, came out 4 days ago — too new to call settled.

Fine for most users; self-hosted Docker upgrades need manual steps, and Docker/codex, web-paste, and model-failover quirks remain open

Why this verdict

v2026.10.5 is a large release (2,704 commits, 2,250 PRs) and the evidence supports updating for most users. The deterministic evidence gate is ✅ — no credible blocking issue is confirmed for this version — and the calibration data backs that up: against an install base of ~2.1M weekly npm downloads, every one of the 60 tracked issues has zero 👍 reactions, with the most-commented report (#179676) drawing only 5 comments. A genuinely widespread breaker draws dozens-to-hundreds of reactions within days; none of these reports show that signature. Upstream velocity is healthy too, with 25 of 85 ever-tracked issues already closed as fixed and none closed as wontfix. This matches the prior assessment of this version (✅, 2026-10-21), and the broken-state has not materially worsened since.

Known issues (60+)

• #179676 (high) [Bug]: Control UI pasted text (>1,000 chars) only reaches the model on the turn it is sent; later turns lose it
• #177068 (high) Deep-consolidation LLM budget in memory-core dreaming is hardcoded to 60 s — no config knob, unlike the narrative phases fixed in #167984
• #177077 (high) [Bug] v2026.10.5 still lacks the Chinese quota-exhaustion failover fix from #172173 — MiniMax 429 retries 9x before fallback
• #177078 (high) [Bug]: Quoting a file message on DingTalk fails with "Transcript idempotency key ... conflicts with the admitted message"
• #177210 (high) [Bug]: Webchat continuation turn acts on a stale echoed user request from injected runtime context, producing an unrequested duplicate answer
• #177387 (high) [Bug]: Bundled MiniMax catalog keys M3.1 on the "MiniMax-M3.1-Flash-Preview" id only, so the documented "MiniMax-M3.1-Flash" id silently loses the 5-level thinking profile, the output_config stream patch, and all M3.1 defaults
• #177909 (high) [Bug]: transcript idempotency conflict is treated as a model failure and walks the cross-provider fallback chain (2026.10.5)
• #178804 (high) [Bug]: Codex plugin: app-server processes accumulate until container hits pids limit (pthread_create EAGAIN), gateway becomes unusable; plugin is auto-re-enabled by OpenAI ChatGPT route

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.