Gemini 3.1 Pro Service Anomaly Report

Gemini 3.1 Pro Regional Outage: Asia-Pacific and Western U.S. Users Face Total Failure

Over the past 6 hours, Gemini 3.1 Pro experienced large-scale service anomalies in Asia-Pacific and the western and central United States. Users in multiple locations reported error rates of 100% and average latency dropping to zero, while service in the eastern United States remained normal. Three key reports on X indicated that the issue was concentrated between 15:25 and 15:40 UTC, pointing to an evident regional deployment or routing failure.

User reports were highly consistent. Asia-Pacific users were the first to report an "LLM Outage," followed in succession by users in the western and central United States experiencing the same symptoms. Some developers tried to route around failed nodes, but most calls still returned failures. This corroborated the real-time monitoring results of the YZ Index: the model's main leaderboard score plummeted from 77.7 to 65.92, a drop of more than 11 points. Targeted retesting (enhanced 18-question version) further revealed an execution score of only 57.10, far below the solidity score of 76.70, indicating that even on some available nodes, the model's instruction-following and multi-step reasoning capabilities had significantly degraded.

From a technical perspective, this incident most likely resulted from an anomaly in regional model deployment or load-balancing policies. Gemini 3.1 Pro uses a multi-region distributed architecture. The Asia-Pacific and western U.S. nodes are suspected to have triggered a hot update or configuration push of the same version within the same time window, causing the inference service to fail to load weights or establish sessions normally. The fact that US-East remained normal ruled out the possibility of global model parameter corruption, pointing more toward local configuration drift or failure of internal network policies in data centers.

In addition, the sharp drop in execution score suggests that some requests may have been downgraded to an older version or a lightweight proxy model, further amplifying the perceived performance cliff among users. In the short term, Google needs to prioritize checking version consistency and routing policies for Asia-Pacific and western U.S. nodes, and disclose the root cause to restore developer confidence.

This outage once again exposes the fragility of multi-region large-model services in canary releases and real-time monitoring. For production systems that rely on Gemini 3.1 Pro, it is recommended to immediately enable cross-region fallback and multi-model parallel invocation strategies.