Severe latency & low throughput (~20 tps, high TTFT) on Gemini 3.8 Flash over past 24h (likely upstream Google issue)

Problem Description

Over the past 24+ hours, requests using Google Gemini 3.8 Flash have suffered severe performance degradation.

Observed Metrics

• Generation Throughput: Dropped to roughly ~20 tokens per second.
• Time to First Token (TTFT): Unusually high latency / noticeable stall before the first token or chunk is returned.
• Timeline: Ongoing degradation for approximately 24 hours.

Note

Given that this degradation is occurring consistently across different clients, tools, and proxy layers, the root cause appears to be upstream on Google's infrastructure/API side rather than a defect in this repository. Opening this issue to document the ongoing slowdown, track upstream resolution, and see if others are observing the same behavior.