GPT Models takes too long to respond

Description

So I tried simple prompting using GPT models sometimes it able to respond within seconds but sometimes it takes several minutes even though it is a very simple command. for example I tried GPT 5.4 with a variant of xhigh and I just tried to update the graphify but it takes too long of the AI to respond or finish the execution.

Plugins

graphify

OpenCode version

1.15.10

Steps to reproduce

choose any GPT models and maybe like high or xhigh variant

Screenshot and/or share link

it is already about 10 minutes now

Operating System

endevourOS arch linux

Terminal

bash

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

GPT-5.4 is unresponsive on first prompt, but starts working after switching to 5.3 Codex and back #16870: GPT-5.4 is unresponsive on first prompt — similar symptom of GPT models taking too long or failing to respond

If your issue is different or not covered by the above, please add more details to help distinguish it.

I checked on the https://status.openai.com/ it is operational

+1, same here. There seems to be an issue with GPT-5.4. GPT-5.5 is somewhat better

Same, looks like OpenAI is cooking something for 3rd party harness tools. AI doom is coming

+1, same here. There seems to be an issue with GPT-5.4. GPT-5.5 is somewhat better

for me a simple prompting like change files names it takes longer for both GPT 5.4 or GPT 5.5

+1, same here. There seems to be an issue with GPT-5.4. GPT-5.5 is somewhat better

for me a simple prompting like change files names it takes longer for both GPT 5.4 or GPT 5.5

Same. I think it's due to new update. I rolled back two versions and it works fine.

@AmanGusain2003

Same over here @AmanGusain2003 which version are u using?

Btw, I checked the status page and its operational

@AmanGusain2003

Same over here @AmanGusain2003 which version are u using?

Btw, I checked the status page and its operational

Right now I'm using v1.15.6. But I'm again having issues, like one off replies are now faster but agentic mode is getting stuckk. This is killing me. I don't know if it's an issue with opencode, ohmyopencode or in general GPT models.

And when I interrupt it snd then send "." It works

this needs to be resolved ASAP

@jlongster

@jlongster

Issue is present in v1.14.50 too,

this still happens with opencode 1.15.10 :(

This is almost certainly a provider serving issue in SEA/Oceania. I'm currently in Philippines and having my European VPN on magically fixes the stuck and hanged states on OpenAI Provider and my flows work normally. But OpenCode not having valid paths to get these states retried after an adequate time doesn't help for sure.

Note: I run my own fork and I have just now updated it from being a month old relative to upstream to actual, this is most certainly not a regression on the part of OpenCode since the issue is present only for the last 2 days and doesn't seem to depend on the OpenCode version.

Yeah, seems like it is an issue with provider. But certainly we can have better retry mechanism as codex can kind of brute force through this. And opencode just kinda hangs in there.

This is almost certainly a provider serving issue in SEA/Oceania. I'm currently in Philippines and having my European VPN on magically fixes the stuck and hanged states on OpenAI Provider and my flows work normally. But OpenCode not having valid paths to get these states retried after an adequate time doesn't help for sure.

Note: I run my own fork and I have just now updated it from being a month old relative to upstream to actual, this is most certainly not a regression on the part of OpenCode since the issue is present only for the last 2 days and doesn't seem to depend on the OpenCode version.

If anyone is still facing this issue that isnt using websockets openai requested the following info to help them trace down issue (I have no way of replicating this bug unfortunately):

could you capture one current reproduction on the path that is still hanging, with the 10s retry relaxed just for that controlled test? What would help us is:
model and transport
request start timestamp
whether/when response headers arrive
cf-ray if headers arrive
whether/when the first stream event arrives
No prompt or output needed. This should let us distinguish a headers/connect issue from a first-event stall.

Easiest way to do this is run mitm proxy and set the HTTPS_PROXY vars. For example I have this zsh function I wrote so i can do proxy opencode:

proxy () {
        if [[ -z "$1" ]]
        then
                echo "Usage: proxy <command>"
                return 1
        fi
        mitmweb --listen-host 127.0.0.1 --listen-port 58888 --web-open-browser &> /dev/null &
        local pid=$!
        echo -n "Waiting for mitmweb..."
        while ! nc -z 127.0.0.1 58888 2> /dev/null
        do
                sleep 0.2
        done
        echo " ready"
        export NODE_EXTRA_CA_CERTS="/Users/$USER/.mitmproxy/mitmproxy-ca-cert.pem"
        export HTTPS_PROXY="http://127.0.0.1:58888"
        export HTTP_PROXY="http://127.0.0.1:58888"
        export NO_PROXY="localhost,127.0.0.1"
        "$@"
        kill "$pid" 2> /dev/null
        wait "$pid" 2> /dev/null
        unset NODE_EXTRA_CA_CERTS HTTPS_PROXY HTTP_PROXY NO_PROXY
}

@rekram1-node

I wrapped OpenCode in mitmproxy to investigate as to what I can see is happening.

Unfortunately, it looks like requests are made to https://chatgpt.com/backend-api/codex, but no response is ever received from OpenAI. After ten minutes, OpenCode disconnects and retries the HTTP 1.1 call and it usually continues (usually gets routed to a new TCP/TLS connection server as well):
Timing:
First request byte:	2026-05-25 14:24:10.323
Request complete:	2026-05-25 14:24:10.333(10ms)

Client disconnected.
2026-05-25 14:34:10.340

@rekram1-node it was done by @OverHash

For what it is worth: have been running 1.15.12 for about 24 hours (in server/attach mode, with websockets), no issues, performs well. Can't yet say if it is as fast as codex again, but it is definitely responsive.

Ill have another websocket update soon that should have parity w/ codex (taking advantage of the previous_response_id logic to reduce latency)

Is it just me or is GPT 5.5 just not working at all in Opencode.

An error occurred while processing your request. You can retry your request, or ... (click to expand) [retrying in 2s attempt #2]

I have tried adding env var above. I have tried removing it. I have tried VPN . I have tried upgrade to dev build.

Codex works fine but it aint OpenCode

@thaoula

@thaoula are u using a ton of mcp servers by chance? or any plugins?

This is all I am using.

I think the mcp servers are eating the entire context window or ur supplying too many tools, they dont really give a good error in this case, we will add a tool search or code mode soon to resolve, Im assuming if u disable all the mcps it will work again, or maybe just 1 of them is causing the issue, hard to say

It was the MCP servers. 1.15.13 is working great. Many thanks 🙏🏻

What are the tool limits and should they be tracked by OpenCode on a model by model basis so you can give clear warning and not rely on model responses? I know Grok Build will error if you specify more the 200 tools.

Im doing an mcp pass this week I hope to have better state of things by eow

I am still occasionally seeing an interruption issue with the dev build while using the experimental WebSocket path.

Environment:

OpenCode version: @dev (0.0.0-dev-202606011039)
OS: Windows
Started through a .bat file with proxy environment variables
OPENCODE_EXPERIMENTAL_WEBSOCKETS=true

Batch file:

@echo off
set NO_PROXY=localhost,127.0.0.1
set ALL_PROXY=http://127.0.0.1:18888
set HTTP_PROXY=http://127.0.0.1:18888
set HTTPS_PROXY=http://127.0.0.1:18888
set OPENCODE_EXPERIMENTAL_WEBSOCKETS=true
opencode %*

Observed behavior:

GPT usually starts and responds normally.
The issue is intermittent and not very frequent.
However, on long-running YOLO-style tasks, OpenCode may eventually show interrupted.
Because of this, it is hard to leave GPT running unattended for a long time.

do i have to reconnect codex?

Hi @rekram1-nodeI just wanted to ask for an update: WebSocket support is currently available only in the beta version, right? It is not available in the stable version yet, correct?

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

GPT Models takes too long to respond #29079

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

96 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

GPT Models takes too long to respond #29079

Description

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

github-actions commented on May 24, 2026

MarcoLouisNarca commented on May 24, 2026

mrjared16 commented on May 24, 2026

spidynots commented on May 24, 2026

MarcoLouisNarca commented on May 24, 2026

AmanGusain2003 commented on May 24, 2026

LuisUrrutia commented on May 24, 2026

AmanGusain2003 commented on May 24, 2026

MarcoLouisNarca commented on May 24, 2026

MarcoLouisNarca commented on May 24, 2026

ikarsokolov commented on May 24, 2026

LPFchan commented on May 24, 2026

glyphloom commented on May 24, 2026

AmanGusain2003 commented on May 24, 2026

96 remaining items

rekram1-node commented on May 29, 2026

kunal-004 commented on May 29, 2026

berenddeboer commented on May 30, 2026

rekram1-node commented on May 30, 2026

thaoula commented on May 30, 2026

rekram1-node commented on May 30, 2026

thaoula commented on May 30, 2026

rekram1-node commented on May 30, 2026

thaoula commented on Jun 1, 2026

rekram1-node commented on Jun 1, 2026

ThisCCL commented on Jun 2, 2026

smashah commented on Jun 17, 2026

MarcoLouisNarca commented on Jun 30, 2026

github-actions commented on Aug 29, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions