OpenClaw 2026.4.14: Smarter GPT-5.4 Routing, Chrome CDP Upgrades, and Messaging Reliability Fixes

OpenClaw released version 2026.4.14 on April 14, 2026, marking another significant quality release focused on model provider reliability, browser automation, and messaging platform fixes.

GPT-5.4 Smarter Routing and Recovery

The headline feature is improved routing for GPT-5 family models. The release adds forward-compatibility for gpt-5.4-pro, including Codex pricing and limits, with list/status visibility before the upstream catalog catches up. The routing improvements are designed to dynamically select and fail over between available model providers, reducing task latency and error cascades in multi-agent pipelines.

Chrome and CDP Improvements

Chrome and Chrome DevTools Protocol (CDP) functionalities received meaningful upgrades, improving browser automation stability for data extraction and RPA use cases. The CDP enhancements make OpenClaw’s web interaction capabilities more reliable for production workflows.

Subagent Fixes and Messaging Platform Improvements

Subagents that previously could get stuck in loops received attention in this release. Cross-platform messaging improvements cover Slack, Telegram, and Discord — including Slack interactive event allowlisting hardening and Telegram forum topic name surfacing in agent context.

Security Hardening

Security fixes include closing a gateway-tool configuration bypass vulnerability (preventing config.patch/config.apply calls that would enable security-audit-flagged options like dangerouslyDisableDeviceAuth), and a media-attachment path canonization fix that prevents realpath errors from downgrading allowlist checks.

Full Changelog Highlights

  • OpenAI Codex/models: Forward-compat for gpt-5.4-pro before upstream catalog sync (#66453) — thanks @jepson-liu
  • Telegram/forum topics: Surface human forum topic names in agent context and plugin hook metadata by learning from Telegram service messages (#65973) — thanks @ptahdunbar
  • Agents/Ollama: Configured embedded-run timeout now properly forwarded to undici stream timeout tuning, fixing slow local Ollama runs getting cut off at default stream cutoff instead of operator-set run timeout (#63175) — thanks @mindcraftreader and @vincentkoc
  • Models/Codex: Fixed apiKey missing from codex provider catalog output, which was causing Pi ModelRegistry validator to silently drop all custom models (#66180) — thanks @hoyyeva
  • Tools/image+pdf: Normalize provider/model refs before media-tool registry lookup — Ollama vision models no longer incorrectly rejected as unknown (#59943) — thanks @yqli2420 and @vincentkoc
  • Slack/interactions: Apply global allowFrom owner allowlist to block-action and modal interactive events; require expected sender ID for cross-verification; reject ambiguous channel types (#66028) — thanks @eleqtrizit
  • Media/attachments: Fail closed on unresolved local attachment paths via realpath; network fetch fallback preserved for attachments that also have URLs (#66022) — thanks @eleqtrizit

Operators running OpenClaw should upgrade to 2026.4.14 to take advantage of the reliability improvements.

← Back to News