OpenClaw v2026.4.14 Adds GPT-5.4-pro Support and Fixes Ollama Timeout Forwarding

OpenClaw v2026.4.14 now supports GPT-5.4-pro and fixes a timeout bug that could interrupt slow Ollama runs before they finished streaming a response. The release, published by Petronellatech on April 14, 2026, is aimed at people running AI agents that mix hosted models with local ones, including users who rely on Ollama for on-premises inference.

The new GPT-5.4-pro support is described as forward-compatible, which usually means the software has been updated to recognize and work with a newer model family without requiring a major code change in the application itself. For users, that kind of support matters because AI agent platforms often need to keep pace with model launches from providers like OpenAI while still preserving the same agent workflows, tool calls, and message formats.

The more practical fix in this release concerns Ollama, the open-source local model runner that many teams use to keep model execution on their own machines or internal servers. In earlier versions of OpenClaw, if a local model took too long to reason through a request, the request could hit the platform’s global stream timeout and disconnect before the model had finished sending its output.

That behavior is especially frustrating for long reasoning tasks, because local models often trade raw speed for privacy, cost control, or offline operation. When an agent is asked to plan, synthesize multiple steps, or work through a complex prompt, the response stream can stay active for longer than a simple chat reply. If the timeout is applied too broadly, the application may drop the connection even though the model is still working normally.

Forwarding the timeout correctly means OpenClaw can treat Ollama jobs differently from faster, remote model calls. Instead of letting the shared streaming limit terminate a slow local inference run, the agent can continue receiving tokens until the local process finishes or until the local request itself ends. That is the kind of plumbing change that users may not notice when it works, but they notice immediately when it fails.

OpenClaw is part of a broader class of AI orchestration tools that sit between models, prompts, tools, and external systems. These platforms are increasingly expected to support both cloud-hosted frontier models and self-hosted alternatives, because teams often want the speed and capability of remote models for some tasks and the control of local models for others.

The combination of a new model compatibility update and a timeout fix also reflects a common reality in agent software: support for the latest hosted model is only useful if the platform’s local execution path is stable enough to handle slower, more deliberate workloads. In practice, that means release work often has to cover both the visible parts of the product, like model selection, and the less visible parts, like streaming behavior and request lifecycle handling.

According to the release note summary, the Ollama fix specifically addresses the case where complex reasoning tasks on local models no longer disconnect mid-response. The update lands in v2026.4.14, alongside the GPT-5.4-pro support.