Release info

Four OpenClaw Releases in Eight Days: GPT-6 Astra, Plugin Workspace and Update Rollback

The OpenClaw 2026.9 stable line shipped four releases in eight days: v2026.9.1 (Mermaid rendering, quick start), v2026.9.2 (GPT-6 Astra), v2026.9.3 (update rehearsal, Node 24.16+ required) and v2026.9.4 (unified Plugins workspace, safer automatic rollback, prepared cloud sessions). Pi 0.85.x added GPT-6 Astra; DeepSeek Harness shipped 0.1.5-rc.1.

Direct answer: The OpenClaw 2026.9 stable line shipped four releases in eight days — v2026.9.1 (Sep 3: Mermaid rendering, quick-start lane, personal skill libraries), v2026.9.2 (Sep 5: GPT-6 Astra), v2026.9.3 (Sep 8: updates rehearsed in isolation; Node 24.16+ now required) and v2026.9.4 (Sep 11: unified Plugins workspace, automatic rollback for compatible failed updates, prepared cloud sessions). Pi added GPT-6 Astra in 0.85.1; DeepSeek Harness shipped 0.1.5-rc.1 with V4.1-Flash support.

OpenClaw v2026.9.1: the September line goes stable

Official release highlights (stable channel):

  • Diagrams in every chat: Mermaid blocks render as diagrams in the Control UI and the native macOS, iOS and Android apps, with enlarge previews and a mobile retry on render failure;
  • From install to chat in one prompt: fresh installs (including npx openclaw@latest) enter a quick-start lane that detects existing Claude Code or Codex logins and API keys, verifies them live and opens the web dashboard from a foreground Gateway; the full wizard remains as Custom setup;
  • Personal skill libraries on shared Gateways: openclaw skills library gives everyone a skill set separate from the workspace, with ZIP import and per-identity sharing or publishing on team Gateways;
  • Updates that leave you working: openclaw update rolls back the npm candidate when the post-update Doctor fails, preserves configuration and secret references across failed upgrades, and hands failures to built-in triage.

The official stable release was published on September 3. The site's upgrade tutorial covers the key changes in the 2026.9 line.

v2026.9.2 (Sep 5): GPT-6 Astra arrives, performance shored up

  • GPT-6 Astra support: openai/gpt-6-astra works with an OpenAI API-key profile or an eligible ChatGPT/Codex account, with text and image input, Responses tool calls and supported reasoning controls;
  • Faster, more responsive chat: chat, dashboards and session interactions stay responsive while long transcripts and disk usage are processed, with durable history reads moved outside the Gateway event loop;
  • Replies survive restarts: active, queued and delegated replies recover after Gateway restarts, with continuation instructions carried through compaction and retries.

v2026.9.3 (Sep 8): rehearse updates first, Node bar raised

  • Safer updates: core and plugin changes are rehearsed in an isolated candidate state before activation; eligible 2026.9.2 migrations are supported and abandoned update records recover without stopping a healthy Gateway;
  • ⚠️ Breaking — Node runtime: Node 24 needs ≥ 24.16.0, or Node 26 ≥ 26.1.0 (Node 26 recommended); Node 22, 25 and older 24.x/26.x builds are no longer supported — upgrade Node BEFORE OpenClaw to avoid SQLite text truncation;
  • Skill Workshop persistence: skills live in one persistent agent-owned collection across workspaces, with full-instruction comparison;
  • Sharing and meetings: publish a revocable read-only link to a session; a searchable meeting library lands in the Control UI (browse notes, full-text transcript search, Markdown/JSONL export);
  • Provider accounts together: manage connected accounts and priority in Models settings.

v2026.9.4 (Sep 11): unified Plugins workspace, dependable rollback

  • Recover from compatible failed updates: when schema and configuration checks prove rollback is safe, the previous package, command shim, service and pre-update configuration are restored; migration-bearing failures still require a verified pre-update backup;
  • Plugins in one place: discover bundled and ClawHub plugins, install from the Control UI, and manage setup and access in a unified Plugins workspace;
  • Prepared cloud sessions: start eligible Linux sessions from local projects or public GitHub repositories, with reusable snapshots built from the Control UI;
  • GPT Image 2.5: pick the new Flare and Sunburst variants for image generation and editing via OpenAI or fal;
  • Read-only config mode: OPENCLAW_CONFIG_READONLY=1 keeps OpenClaw from rewriting deployment-managed configuration.

Also: the June LTS line closed out with v2026.6.35 (Sep 10), the final Extended Stable release for June.

Pi 0.85.x: two versions in two days, GPT-6 Astra arrives

  • v0.85.0 (Sep 4): persistent Claude thinking effort; fullscreen transcript controls; restorable in-memory sessions (SessionManager.inMemory() for externally managed entries via the SDK);
  • v0.85.1 (Sep 5): GPT-6 Astra enters the model picker (OpenAI API keys and OpenAI Codex subscriptions); five-times-faster Alt+wheel scrolling in fullscreen; fixes SDK import failures caused by 0.85.0 accidentally shipping experimental code.

Pi's two-week model streak (DeepSeek V4 Flash Vision last week, GPT-6 Astra this week) keeps proving the kernel route's model agility — see the deep dive and product page.

Meanwhile: DeepSeek Harness 0.1.5-rc.1

After the 0.1.3-alpha line (Sep 4/7), DSH jumped to 0.1.5-alpha (Sep 8/9) and shipped the first 0.1.5 release candidate on Sep 10: a new DeepSeek-V41-Flash (deepseek-flash) adapter (default for new sessions, dynamic system-prompt updates that preserve the KV cache); web uploads of arbitrary file types with a multi-tab right Sidebar preview (Markdown, code, HTML, PDF, images); queueable, steerable subagent messages; rc.2 followed the same day with feedback and file-card polish. Per the official release notes, 0.1.5-rc.1 also upgrades the session data format to V3: supported older logs migrate with their originals preserved, and sessions upgraded to V3 do not support downgrade reads. All three self-hosted leaders moved this week; cross-check the comparison hub.

Sources