Very frequent errors when using openai

Description

Using openai/gpt-5.4, see so many of these:

{"type":"error","sequence_number":2,"error":{"type":"server_error","code":"server_error","message":"An error occurred while processing your request. You can retry your request, or contact us through our help center at help.openai.com if the error persists. Please include the request ID e54d47cb-7060-4038-a6d2-c8d4cef9e303 in your message.","param":null}}

a dozen a day at least. Quite frequenty since maybe the past 3 days? Thinking level doesn't seem to be important.

Plugins

No response

OpenCode version

1.14.21

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

Ubuntu 24.04

Terminal

Alacritty/tmux

This issue might be a duplicate of an existing issue. Please check:

Intermittent OpenAI streamed server_error (sequence_number:2) with gpt-5.3-codex; retries degrade session #16214: Intermittent OpenAI streamed server_error (sequence_number:2) — describes the same error JSON payload from OpenAI's streaming API, intermittently occurring during requests. The main difference is the model version (gpt-5.3-codex vs gpt-5.4), but the underlying symptom appears identical.

I’m experiencing the same issue as well. It has been happening continuously for the past hour on my side.

Same issue here, experiencing this since yesterday when it started happening about 20% of the times. Today is about 95%, making opencode unusable. If it helps I also tried t3code and I had issues there as well. Using openai codex works 100% of the times. Such a pity, cause opencode is far superior than codex (they don't even have an undo command which blows my mind)

Same issue here

I can confirm a very similar server_error pattern on my side when using OpenCode with OpenAI/Codex. The important difference is that I do not see this when using Codex CLI directly with the same account/workflow.

That makes me suspect this may be specific to OpenCode provider integration or streaming/agent handling, rather than a general Codex outage.

My setup:

OS: Windows
OpenCode version: 1.14.20
Provider/model: openai/gpt-5.4
Terminal: Hyper, Windows Terminal

+1 here, infuriating, i'm trying to track down

hm ill look

I have inconclusive medium confidence at best so far, but here are things I believe

It's happening specifically with task().
It happens much more often/consistently with gpt 5.5, but also sometimes with 5.4.
with 5.5 it happens the majorit of the time but not always.
at first I thought it was happening with the general agent but not exlore. But that seems to have been a red herring
a trivial promot sending back a trivial answer doesn't trigger it. has to be substantial.
happens on an older version + on v1.14.33.
I'm using OpenAI API key.

It's actually strange. An hour ago, I felt like I'd already narrowed down the conditions enough that I would be able to find a root cause or at least an independent standalone repo very fast. And I'm quite surprised that I have not.

@rekram1-node

@rekram1-node if you have trouble repro-ing lmk, I will attempt to help.

might have something to do with passing too many tools + invoking a tool call

uh I think it's just their server having outages I think the fix is to just auto retry this I dont think it's some bug on our end

Unless you are saying that once the error occurs the session is permanently broken?

i agree it looks like this at first but i think it's not. i'll report back.

Having the same issue, happens almost always with gpt 5.5, often with gpt 5.4, and never with OpenCode Go models

It’s sending too many tools, not positive yet. You send too many tools and then make a tool call. Mike Harris ***@***.*** (201)650-5579

…

On Tue, May 5, 2026 at 9:46 PM Salloxy ***@***.***> wrote: *Salloxy* left a comment (anomalyco/opencode#23944) <#23944 (comment)> Having the same issue, happens almost always with gpt 5.5, often with gpt 5.4, and never with OpenCode Go models — Reply to this email directly, view it on GitHub <#23944 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAPWKGZ5Y2V66XFO3OAZTL4ZKKNTAVCNFSM6AAAAACYDQCZAOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DGOBUGQ4TEMJTGM> . You are receiving this because you commented.Message ID: ***@***.***>

It’s sending too many tools, not positive yet. You send too many tools and
then make a tool call.

I think this is the issue—disabled one MCP with many tools, and the issue is resolved.

It’s sending too many tools, not positive yet. You send too many tools and
then make a tool call.

I think this is the issue—disabled one MCP with many tools, and the issue is resolved.

I only have 3 mcp servers: exa, chrome-devtools, resend. After disabling them, this issue no longer appears.

It’s sending too many tools, not positive yet. You send too many tools and
then make a tool call.

I think this is the issue—disabled one MCP with many tools, and the issue is resolved.

yep seems to be the same for me

same

Same, disabled one MCP with many tools, and the issue is resolved.

Adding Windows/Desktop v1.17.11 evidence for the same OpenAI streamed server_error family.

Sanitized environment:

Windows 11 x64
OpenCode Desktop v1.17.11
Provider OpenAI via Codex OAuth
Model/variant in session DB: openai/gpt-5.5-fast / xhigh
Runtime: ai-sdk

Observed:

Desktop showed retry UI: An error occurred while processing your request... retrying in 18s - attempt #32
Logs repeatedly showed:

message="stream error" providerID=openai modelID=gpt-5.5-fast
error.error.error.type=server_error
error.error.error.code=server_error
error.error.error.message="An error occurred while processing your request..."

Same session kept retrying under the same run for many attempts; observed sequence numbers included 3, 61, 101, 174, and 222.
The UI continued after some retries, but became extremely slow; in practice gpt-5.5-fast was slower than gpt-5.5/high because of retries.
This started after updating to the June 25 v1.17.11 Desktop build.

Mitigation:

Stop/restart Desktop.
Disable gpt-5.5-fast variants in config.
Force default agent model to openai/gpt-5.5 / high.
A default CLI smoke run then selected gpt-5.5/high and completed.

This looks related to retry classification/handling plus OpenAI Responses stream errors. Even when retried, a session can get into a costly retry loop rather than recovering cleanly.

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

Very frequent errors when using openai #23944

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Very frequent errors when using openai #23944

Description

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

github-actions commented on Apr 23, 2026

induge commented on Apr 23, 2026

JaviOverflow commented on Apr 23, 2026

reimurashiki commented on Apr 23, 2026

mharris717 commented on May 4, 2026

rekram1-node commented on May 4, 2026

mharris717 commented on May 4, 2026

mharris717 commented on May 4, 2026

mharris717 commented on May 4, 2026

rekram1-node commented on May 4, 2026

mharris717 commented on May 4, 2026

Salloxy commented on May 6, 2026

mharris717 commented on May 6, 2026

brennandai commented on May 7, 2026

Salloxy commented on May 7, 2026

allen-moonshot commented on May 7, 2026

ugoi commented on May 21, 2026

ouchxp commented on Jun 2, 2026

NedMalki-Chief commented on Jun 25, 2026

github-actions commented on Aug 25, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions