Causes and solutions for “Message stream error”

“Message stream error” appears when the streaming connection through which ChatGPT receives a response incrementally from the server is interrupted. In most cases, the cause is not user error but server overload, an unstable connection, or a browser-environment problem. Many cases are resolved by regenerating the response or reloading the page. If that does not help, the fastest sequence is to check outage information → isolate the connection and browser → move to a new chat.

The error

Error shown on screen

メッセージストリームでエラーが発生しました

（英語 UI では "Hmm...something seems to have gone wrong." などと
　表示される同種のエラー。回答が途中で止まる・最後まで生成されない
　という症状を伴う）

Causes and fixes at a glance

Cause	Fix
Server-side outage or overload (high traffic or a major incident; some incidents affect all plans simultaneously)	Check outage information at status.openai.com. During an outage, there is nothing the user can do; waiting for recovery is the only response.
Unstable connection (switching Wi-Fi, VPN/proxy use, or weak mobile signal)	Temporarily turn off the VPN or proxy and retry over another connection (Wi-Fi ⇄ mobile).
Browser-environment problem (extension interference, stale cache, or an expired session)	Check whether the issue reproduces in an incognito window. If it does not, disable extensions, clear the cache, and sign in again.
The conversation is too long or attachments are too large (a bloated thread makes response timeouts more likely)	Start a new chat while carrying over the key points. Provide large files in parts.

Three basic steps to try first, within 90 seconds

Regenerate the response → reload the page (if using the app, fully close and restart it) → log out and back in. Cases caused by a temporary connection interruption are resolved by one of these three steps. If regeneration stops at the same place every time, that is a sign of another cause.

Check whether it is an outage rather than a problem on your side

If the same error repeats every few minutes, check status.openai.com (ChatGPT’s official status page) for an active incident. If an outage is shown, changing your settings will not fix it — waiting for recovery will. Searching social media for “message stream” can also show within seconds whether many people are affected.

Isolate the environment one layer at a time

If there is no outage, the cause is in your local environment. Isolate it from the outside in: ① Turn off the VPN and proxy ② Open an incognito window (removes extensions and cache together) ③ Try another browser or device ④ Switch networks (Wi-Fi ⇄ mobile). The step where it starts working again identifies the cause. If it happens frequently in long conversations, carrying only the key points into a new chat is most effective.

For developers: the same phenomenon can occur in the API

When calling the API with streaming (stream: true), the same phenomenon appears as an interrupted SSE connection. An error event can arrive partway through the stream even after HTTP 200 has been returned; under overload, 529 (overloaded_error) or 500 may also be returned. There are three measures: (1) treat a stream that ends without finish_reason as retryable; (2) retry 429, 500, and 529 with exponential backoff and jitter; (3) through a proxy, check the idle timeout and buffering settings. First stream SSE directly without a proxy to isolate where the interruption occurs:

stream-test.sh

# プロキシを介さず SSE を直接確認する（切断の再現テスト）
curl -N https://api.kunavo.com/v1/chat/completions \
  -H "Authorization: Bearer $KUNAVO_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-5","stream":true,
       "max_tokens":300,
       "messages":[{"role":"user","content":"1から20までゆっくり数えて"}]}'

If you’re calling through Kunavo

Kunavo (an AI API gateway) is designed with this kind of temporary streaming failure in mind. When multiple upstream channels are configured for a model, a failed first attempt automatically switches to another channel and retries within the same request, turning a temporary upstream problem into “a slightly slower success” rather than an error. Failed requests are never billed. Because GPT and Claude can be called with one API key, an overloaded model can be handled by changing only the model name.

Frequently asked questions

Is “Message stream error” my fault?

In most cases, no. This error indicates that the connection receiving the response from the server was interrupted; most causes are server-side load or the connection environment. The content you entered generally does not cause this error.

What if it does not go away after repeated regeneration?

Try these in order: ① Check status.openai.com for an outage (if there is one, wait) ② Reopen it in an incognito window on another network ③ If the conversation is long, move only the key points to a new chat. Cases that are not fixed by these three steps are rare; in that situation, waiting a while is the practical option.

Do other AI systems show the same error?

This wording belongs to ChatGPT, but Claude and Gemini use the same mechanism of streaming responses, so an interrupted connection produces a similar error. In Claude it appears as a message that the response could not be fully generated; with direct API calls, it appears as an interrupted SSE stream or 529 (overloaded_error).

Related guides

More error semantics live in the error reference; getting a key takes a minute via signing up and the authentication guide.

“Message stream error” causes and solutions — common to ChatGPT and AI chats