Gemini 2.5 Pro with empty response.text

Gemini API
184

This was happening during rush hour. When I tried using the API during less busy hours, it worked fine.

186

It feels like rush is not causing the issue. Seems like servers are crashing after certain time, and feels like it gets fixed during morning USA time everyday for the past few days. It does not respond at all from 12 pm to 9 pm IST time.

188

Same, itā€™s returning thinking text (which is also cut off) but not the actual response.

1 Like
190

Any updates? @Logan_Kilpatrick

2 Likes
192 
lee@dev-pc "cpp-dev" build %> codegem "This is just a test" ../
[RAW RESPONSE BUFFER]:
[{
  "candidates": [
    {
      "content": {
        "role": "model"
      },
      "finishReason": "STOP",
      "index": 0
    }
  ],
  "usageMetadata": {
    "promptTokenCount": 51,
    "totalTokenCount": 51,
    "promptTokensDetails": [
      {
        "modality": "TEXT",
        "tokenCount": 51
      }
    ]
  },
  "modelVersion": "gemini-2.5-pro",
  "responseId": "sL-laK2DDP7Bz7IPooaNGQ"
}
]
[END RAW RESPONSE BUFFER]

[INFO] The model finished generating but produced no text content.

A simple test even fails Seems like this has been going on for a while?

2 Likes
194

wow weā€™re still here days after and the issue still has not been fixed. All my customer demos have been cancelled. Itā€™s such a shame that Gemini 2.5-pro was the only model that performed reasonably well on my task. Certainly getting a lesson on vendor dependency right here. I will for sure not stay with Gemini for long.

6 Likes
195

HOLY WOW BATMAN. I dont understand how broken this is, and how people are not being fired for this. Releasing it as a paid service when it is so broken.

Now, it completely does not work for me (the original poster). All my little retry measures which worked just fail 100% now and we are rushing to release our product.

I guess it is really time to switch, which is a huge shame.

@Mrinal_Ghosh, can you provide an update other than ā€œwe are working on itā€ and how do we get refunds for this. I have spent a few thousand already on this project assuming Gemini Pro 2.5 was ready for a production solution.

Once upon a time in enterprise software, this would be escalated with almost hourly updates. Sort of miss those days.

8 Likes
196

Running into the same issue with gemini-2.5 flash. Google pushed us to migrate from gemini-1.5-flash to gemini-2.5-flash and this is an ongoing issue since last few months.

2 Likes
197

They have to stop scaling out models, fix your previous one first thatā€™s my advice.

3 Likes
198

When Gemini 2.5 Pro works, its performance is absolutely perfect for our use case. But now the service is unusable. The retry workaround is failing. Weā€™re seeing 100% failure rates. Not a single request can get through even with an aggressive retry policy.

[0] ---
[0] [AIManager] Requesting answer from model: gemini-2.5-pro
[0] [GeminiApiService] API call failed for model gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.
[0] [AIManager] Retryable error for gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.. Retrying... (1/15). Waiting 500ms.
[0] [RATE-LIMITER] Limit (TPM). Waiting: 41897ms.
[0] [GeminiApiService] API call failed for model gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.
[0] [AIManager] Retryable error for gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.. Retrying... (2/15). Waiting 500ms.
[0] [RATE-LIMITER] Limit (TPM). Waiting: 52463ms.
[0] [GeminiApiService] API call failed for model gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.
[0] [AIManager] Retryable error for gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.. Retrying... (3/15). Waiting 500ms.
[0] [RATE-LIMITER] Limit (TPM). Waiting: 53679ms.
[0] [GeminiApiService] API call failed for model gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.
[0] [AIManager] Retryable error for gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.. Retrying... (4/15). Waiting 500ms.
[0] [RATE-LIMITER] Limit (TPM). Waiting: 51092ms.
[0] [DB] Cleaned up 34 old rate limit event records.
[0] [GeminiApiService] API call failed for model gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.
[0] [AIManager] Retryable error for gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.. Retrying... (5/15). Waiting 500ms.
[0] [RATE-LIMITER] Limit (TPM). Waiting: 54162ms.
[0] [GeminiApiService] API call failed for model gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.
[0] [AIManager] Retryable error for gemini-2.5-pro: [GeminiApiService] Received an empty but valid response structure for model gemini-2.5-pro.. Retrying... (6/15). Waiting 500ms.
[0] [RATE-LIMITER] Limit (TPM). Waiting: 47241ms.
[1] vite exited with code SIGINT
[0] nodemon server.js exited with code SIGINT

Apparently, the bug is not easy to fix, as it has been ongoing for such a long time.
While I understand that such issues can be incredibly challenging to resolve, the current state makes the model unsuitable for use in any production system. We are hoping for a swift resolution and more transparent communication from the team.

6 Likes
199

When I use Gemini 2.5 Pro 2 days ago, I always encounter a POST 500 error. However, when I switch to Gemini 1.5 Pro, it works correctly and responds quickly. Iā€™m wondering why the newer version of the API seems to perform worse than a model that was released two years ago??? And I just wonder why the bug still exists even after 48 hours!

201

Guys, same problem for me. Thanks god I discovered now this thread and seen that Iā€™m not the only one.

Iā€™m using Gemini 2.5 Pro for video analysis in client projects, but Iā€™m experiencing failure in 4 out of 5 attempts. This inconsistency is becoming a serious problem for my business.

Hoping to see it fixed ASAP @Logan_Kilpatrick

203

its been like that for like 1-2 weeks still no fix

204

It seems like there was just an update, and the quota limit for 2.5 Pro was reduced from 100 to 50.

1 Like
206

Hi do you have a source for this info? A Google blogpost or news item?

208

You can check on the cloud console

209

By the way, today gemini-2.5-flash worked flawlessly, i did 64 API calls without any error.

210

Sorry but this answer is quite ambiguous. GCC is huge. I can not find any news in the model specs in Vertex AI: Google Cloud console

What information did you find on GCC?

Tnx

212

@Mohamed_Amine I first reported this here in April, so more like 4 months now

214

So, I use GCC in Italian, and Iā€™ll try to guide you as best I can. The fastest way is to go to the Google AI Studio dashboard, under ā€œUsage and Billingā€ at the bottom right, and select ā€œOpen in Cloud Console.ā€ Then go to ā€œQuotes and System Limitsā€ and search for ā€œRequest limit per model per day for a project in the free tierā€ for gemini-2.5-pro. I used Google Translate, so some names may not correspond in English.