For around 1 week, gemini-2.5-flash in locations “europe-west4”, “europe-west1”, “europe-west8”, “europe-north1” and “europe-central2” (at least) just stops generating content mid-sentence, without finishing the task.
The REST response is 200, “finishReason” is “STOP” (so the usual). It happens both in normal and in JSON structured output mode. It happens all across our AI services, without having changed a single comma in the prompts, nor the code that makes the requests.
Is anyone experiencing similar? Any reason or solution?
I somewhat assumed that they are making compute space for gemini-3 and we are dealing with a heavily quantized version of 2.5, but I am desperate because our services are heavily disrupted and we are drowning in support cases.