GPT-5 requests with streaming disabled frequently time out

[jerkstorecaller]

What happened?

TLDR:

• GPT-5 is a "protected" OpenAI model. They don't allow you to use it with response streaming enabled, unless you "verify your organization" (send them a photo of you and your government ID), which I don't want to do. So I'm using GPT as an unverified user.
• To use GPT-5, you must set disableStreaming=true in the model's config in librechat.yaml
• With streaming disabled, on large prompts/replies, I usually (but not always) get this error after a while. I think it's a hidden timeout.

2025-08-24 21:45:12 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted
2025-08-24 21:49:59 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted
2025-08-24 21:53:12 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted
2025-08-24 21:54:40 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted

Clicking Regenerate might make it work, about half the time.

Here's my librechat.yaml, on v0.8.0-rc2:

# Configuration version (required)
version: 1.2.1

# Cache settings: Set to true to enable caching, for better performance
cache: true

modelSpecs:
  enforce: false
  prioritize: true
  list:
    - name: "claude-opus-4.1"
      label: "Anthropic/Claude Opus 4.1"
      description: "Anthropic's flagship model. Use this one if you're not sure."
      default: true
      preset:
        endpoint: "MyOpenRouter"
        model: "anthropic/claude-opus-4.1"
    - name: "gpt-5"
      label: "OpenAI/GPT-5"
      description: "OpenAI's flagship model. Cannot stream response."
      preset:
        endpoint: "OpenAI"
        model: "gpt-5"
        #required on protected OpenAI models
        disableStreaming: true
    - name: "gpt-5-mini"
      label: "OpenAI/GPT-5 Mini"
      preset:
        endpoint: "OpenAI"
        model: "gpt-5-mini"
        disableStreaming: true
    - name: "gpt-4o"
      label: "OpenAI/GPT 4o"
      preset:
        endpoint: "MyOpenRouter"
        model: "openai/gpt-4o"
    - name: "gemini-2.5"
      label: "Google/Gemini 2.5 Pro Preview"
      description: "Google's flagship, reasoning model"
      preset:
        endpoint: "MyOpenRouter"
        model: "google/gemini-2.5-pro-preview"
    - name: "gemini-2.5-online"
      label: "Google/Gemini 2.5 Pro Preview w/ online search"
      description: "Google's flagship, reasoning model, with web search"
      preset:
        endpoint: "MyOpenRouter"
        model: "google/gemini-2.5-pro-preview:online"
    - name: "grok-4"
      label: "xAI/Grok 4"
      description: "xAI's flagship"
      preset:
        endpoint: "MyOpenRouter"
        model: "x-ai/grok-4"



endpoints:
  custom:

    - name: "MyOpenRouter"
      apiKey: "${OPENROUTER_KEY}"
      baseURL: "https://openrouter.ai/api/v1"
      models:
        default: [
          "fakecompany/fakemodelxxx",
          "anthropic/claude-opus-4.1",
          "openai/gpt-4o",
          "openai/gpt-4.1",
          "google/gemini-2.5-pro-preview",
          "google/gemini-2.5-pro-preview:online",
          "x-ai/grok-4"
        ]
        fetch: false
      titleConvo: true
      titleModel: "mistralai/magistral-small-2506"
      forcePrompt: false
      # Recommended: Drop the stop parameter from the request as Openrouter models use a variety of stop tokens.
      dropParams: ["stop"]
      modelDisplayLabel: "OpenRouter"

      # OpenAI entry
    - name: "OpenAI"
      apiKey: "${OPENAI_API_KEY_CUSTOM}"
      baseURL: "https://api.openai.com/v1"
      models:
        default: [
          "gpt-5",
          "gpt-5-mini",
          "gpt-4.1",
          "gpt-4o"
        ]
        fetch: false
      titleConvo: true
      titleModel: "gpt-3.5-turbo"
      summarize: false
      summaryModel: "gpt-3.5-turbo"
      forcePrompt: false
      modelDisplayLabel: "OpenAI"
      disableStreaming: true

#Hide "Code Interpreter" new button in chat input box
interface:
  runCode: false

Version Information

ghcr.io/danny-avila/librechat-dev latest b305361833cb 3 days ago 1.55GB
ghcr.io/danny-avila/librechat-rag-api-dev-lite latest 52622bb7f4fb 7 days ago 1.59GB

Steps to Reproduce

1. In .env, set OPENAI_API_KEY_CUSTOM=valid-api-key
2. Use the above librechat.yaml
3. Try a large request with GPT-5. You can try this prompt:

Your task is to write a 5000 word thesis based on the following abstract:

This thesis presents an examination of the Franco-Ottoman alliance of 1525 within the context of European diplomacy in the sixteenth century. As such, it deals with an event that has often been misunderstood or ignored, and attempts to assign it a place in modern historical understanding of that period. The thesis explores the development of the alliance and the direct connection between the two principals, France and the Ottoman empire, and then examines the effects of the alliance upon politics and diplomacy elsewhere in Europe at the time.

Historical accounts of the alliance to this point are inconsistent. Christian and liberal influences have badly distorted reporting of both the Ottoman empire and the empire's relations with western Europe. Traditional historical accounts of the alliance depict Francis I of France "calling in" the Ottomans to assist him in his struggle with the Holy Roman Emperor Charles V, and then attempting to repudiate the Ottomans when he no longer needed their assistance. Many writers are at a loss to understand how the Christian King Francis could have ever seriously undertaken an alliance with the infidel Ottoman. Despite his supposed reluctance to continue t he alliance, Francis is also widely blamed for Ottoman invasions of eastern Europe, particularly the campaign of 1526 and the defeat of Hungary.

This false moralizing on Francis I and the nature of his Ottoman alliance has, as has been stated, produced a false picture of the alliance and its consequences. It is the purpose of this thesis to provide a new philosophical approach to the Franco-Ottoman alliance, one which will allow for a more objective approach to the alliance and its position in European diplomacy. The formulation of such an approach, a hypothesis concerning the means and ends of the alliance, is viewed as necessary before any further research on the subject takes place. The thesis provides first a chronology of the alliance from the first contacts of 1525-26 to the temporary decline of diplomatic relations during the French Wars of Religion. During this period, Francis I and Henry II of France pursued an active Ottoman alliance as a key part of their foreign policy. After setting out the course of events, the thesis then examines the effects of the alliance in the areas of Europe most affected by combined Franco-Ottoman actions: eastern Europe, including Hungary, the Mediterranean, Venice and the Papacy, and the Habsburg empire of Charles V and his son Philip II. Particular attention has been paid to timing, the exact dates of treaties, campaigns, and embassies, where earlier oversights and errors have led to many misunderstandings of events.

What browsers are you seeing the problem on?

Firefox

Relevant log output

2025-08-24 21:54:40 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

I think it's a hidden timeout.

There is no hidden timeout as defined by LibreChat.

However, some HTTP mechanisms will timeout a request by default if there is inactivity.

To rule some things out, do you have NGINX configured?

See: #4111

What does your setup look like?

[jerkstorecaller]

That was the issue, thanks.
I used an nginx image that defaulted to 90s timeout. I bumped them up to 300s, no more issues.

For anyone finding this through google, these are the settings I changed:

proxy_send_timeout 300s;
proxy_read_timeout 300s;

[danny-avila]

thanks for confirming!