Independent OpenClaw reporting, releases, guides, and community coverage
Releases

OpenClaw v2026.5.12-beta.6: Gateway v4 Protocol, Copilot Image Fixes, and Cron Inspection API

OpenClaw beta.6 ships the Gateway v4 protocol requirement, Copilot OAuth token exchange for image requests, a new cron.get API, and a dozen targeted fixes.

Filed under Releases 4 min read Updated May 14, 2026
Cody
Cody
OpenClaw v2026.5.12-beta.6: Gateway v4 Protocol, Copilot Image Fixes, and Cron Inspection API

OpenClaw shipped v2026.5.12-beta.6 tonight at 21:00 UTC — the sixth beta in the current release cycle and the most wide-ranging yet. It touches the Gateway protocol layer, provider integrations, iMessage, cron tooling, and agent orchestration limits. Here's what changed.

Gateway Protocol v4 Is Now Required

The headline infrastructure change: the Gateway now requires v4 clients (#80725, by @samzong) and streams explicit deltaText and replace frames on chat sessions. This means SDK clients can consume assistant content updates as they arrive — without implementing local text-diffing logic.

If you're building on the OpenClaw SDK, this is the change to act on before the stable release lands. The old implicit-diff behavior is gone; clients that haven't adopted v4 will need to update.

GitHub Copilot: Image Understanding Fixed

A meaningful fix for anyone using OpenClaw with GitHub Copilot: the gateway now exchanges OAuth tokens for Copilot API tokens on image understanding requests and routes Gemini image payloads correctly through Chat Completions. The practical result: Copilot-backed Gemini image descriptions now work correctly after previously silently failing.

iMessage: No More <media:image> Placeholder Text

A community-reported bug (#81209, by @homer-byte): iMessage was sending the literal string <media:image> as visible text when the agent sent media-only messages. The fix preserves the internal echo-deduplication key while stripping the placeholder from outbound content — so image-only sends now deliver cleanly.

New Cron Inspection API: cron.get

Operators running scheduled agent workflows get a new tool: cron.get (#75117, by @samzong). You can now inspect a single stored cron job by ID via:

  • cron.get agent tool
  • openclaw cron get <id> CLI command
  • Direct tool call from inside an agent session

This fills an obvious gap — previously you could list cron jobs but couldn't easily drill into one without parsing the full list output.

Agent-to-Agent Sessions Pre-Created

A reliability fix for multi-agent setups: configured agent main sessions are now created before the first sessions_send or gateway send. Previously, if Agent A tried to message Agent B before B had started a session, the send would fail. Beta.6 removes that ordering requirement — agents can now receive messages without having been active first.

maxPingPongTurns Raised to 20

session.agentToAgent.maxPingPongTurns can now be configured up to 20 (#52400, by @thirumaleshp), up from the previous hardcap of 5. The default remains 5, so existing installations are unaffected — but complex agent-to-agent workflows that needed longer back-and-forth chains can now configure accordingly.

Gateway Now Honors max_completion_tokens

The Gateway's OpenAI-compatible HTTP endpoint now properly forwards max_completion_tokens and max_tokens from inbound /v1/chat/completions requests to upstream providers via streamParams.maxTokens. When both are sent, max_completion_tokens takes precedence. Contributed by @Lellansin — a fix that matters for any SDK or client that sends token budget hints.

OpenAI CLI Auth Default Changes

openclaw models auth login --provider openai now starts the ChatGPT/Codex account login flow by default instead of prompting for an API key. The explicit API-key path remains available via --method api-key. This aligns the default with how most new users are authenticating.

Gemini 3.1 Pro Preview: ID Normalization Complete

A comprehensive sweep across every auth code path — SDK OAuth, direct auth login --set-default, API-key onboarding, per-agent config, proxy-prefixed catalog rows — now consistently emits google/gemini-3.1-pro-preview for Gemini 3.1 testing. No more manually correcting model IDs after auth flows.

fal Provider: Image Edit Mode Improvements

GPT Image 2 and Nano Banana 2 reference-image edit requests are now correctly routed to /edit with an image_urls array (#77295, by @leoge007). Input-image caps are lifted to 10 for GPT Image 2 and 14 for Nano Banana 2. Aspect-ratio hints are now supported in edit mode.

Control UI Recovery Panel

A fallback for blank dashboard pages: when the app module never registers (usually a browser-extension conflict or stale cache), the Control UI now shows a plain HTML recovery panel with a retry path and troubleshooting link. Fixes a longstanding #44107.

Upgrading

v2026.5.12-beta.6 is a pre-release. Install or update with:

npm install -g openclaw@latest

Full release notes are on GitHub. If you're running the beta channel and building on the OpenClaw SDK, the Gateway v4 protocol requirement is the change to prioritize before the stable release.

Daily Briefing

Get the Open-Source Briefing

The stories that matter, delivered to your inbox every morning. Free, no spam, unsubscribe anytime.

Join 45,000+ developers. No spam. Unsubscribe anytime.