You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I don't know what's causing it, but it seems to be coming from the session itself timing out when using the "writing-plans" skill.
OS: macOS Tahoe 26.5 (recently updated)
CPU: Apple M4
More info: "The upstream connection between the client and the model service idled out. This is an infrastructure timeout, not a model error. The model had finished gathering context and was about to produce output, but the connection dropped before the response could be transmitted."
Is this a server error for the model on OpenCode GO?
Thanks for updating the issue! The OS info, error context, and details about the writing-plans skill are helpful. Your report now meets our contributing guidelines.
I also have the same issue with MiMo v2.5 Pro on openrouter. It has been happening for weeks now. Getting it to write large plans always triggers the timeout.
Fedora Linux 44, KDE edition
Reacted by Vitaliy Snitovets, Łukasz Migdałek, VENAXIS, Antar Basil, Nathan (navi), Renan Oliveira, atomic potato, InfHorus, jj1st and Julian W.
same issue but only with Nemotron 3 Ultra Free High. Mimo v2.5 free still okay. current workaround is asking them to write in smaller batches. v1.16.2. using ubuntu linux. terminal alacraty
same issue but only with Nemotron 3 Ultra Free High. Mimo v2.5 free still okay. current workaround is asking them to write in smaller batches. v1.16.2. using ubuntu linux. terminal alacraty
This trick worked for me "small batch writing" for Nemotron 3 Ultra.
same issue but only with Nemotron 3 Ultra Free High. Mimo v2.5 free still okay. current workaround is asking them to write in smaller batches. v1.16.2. using ubuntu linux. terminal alacraty
This trick worked for me "small batch writing" for Nemotron 3 Ultra.
Is not an actual fix but the prompt "Do the edits in small batches" does work as you mentioned, thanks!
I have the same issue, basically OpenRouter is unusable, never seen such unreliable provider, tried 10 differents models they all hit this very same error, I think for a "Router" they should at least make it usable for anything else than sending "hello world" to models.
Activity
[-]"Upstream idle timeout exceeded"[/-][+][BUG] "Upstream idle timeout exceeded"[/+]github-actions commentedon May 23, 2026
Thanks for updating the issue! The OS info, error context, and details about the writing-plans skill are helpful. Your report now meets our contributing guidelines.
RoyaltyInTraining commentedon May 28, 2026
I also have the same issue with MiMo v2.5 Pro on openrouter. It has been happening for weeks now. Getting it to write large plans always triggers the timeout.
Fedora Linux 44, KDE edition
Wydady13 commentedon Jun 8, 2026
i also have same issue with Nemotron 3 Ultra Free High
v1.16.2
VincenzoUrso commentedon Jun 8, 2026
confirm the issue with Nemotron 3 Ultra Free High
ranaharoon3222 commentedon Jun 10, 2026
i also have same issue with mimo pro
joshuac-dev commentedon Jun 10, 2026
I am having the same issue with Qwen Coder Next
cryptodappdeveloper commentedon Jun 10, 2026
All Nemotron models i tried always ended up with "Upstream idle timeout exceeded" when the model trying to use "Writing command"
aalzubidy commentedon Jun 11, 2026
Same issue with DS4 Flash/Pro
leonardocordeiro commentedon Jun 12, 2026
same issue with Nemotron 3 Ultra Free High here
rashidiansanawi commentedon Jun 12, 2026
same issue but only with Nemotron 3 Ultra Free High. Mimo v2.5 free still okay. current workaround is asking them to write in smaller batches. v1.16.2. using ubuntu linux. terminal alacraty
mansoorahmadhh commentedon Jun 13, 2026
I have this Issue with MiniMax M3 Opencode {"code":504,"message":"Upstream idle timeout exceeded","metadata":{"error_type":"timeout"}}
OpenCode 1.17.4 - Macbook Pro Mx
steinathan commentedon Jun 13, 2026
having the issue with
MiMo V2.5namcho commentedon Jun 14, 2026
This trick worked for me "small batch writing" for Nemotron 3 Ultra.
Germanized commentedon Jun 15, 2026
$ErrorActionPreference = "Stop"
$chunkTimeoutMs = 2147483647
$stamp = Get-Date -Format "yyyyMMdd-HHmmss"
Close OpenCode processes if any are still alive.
Get-Process -ErrorAction SilentlyContinue | Where-Object {
$.ProcessName -like "opencode" -or
($(try { $.Path } catch { "" }) -like "opencode")
} | Stop-Process -Force -ErrorAction SilentlyContinue
Start-Sleep -Seconds 2
Reset the bloated Desktop GUI store, but keep a backup.
$desktopDir = Join-Path $env:APPDATA "ai.opencode.desktop"
$store = Join-Path $desktopDir "opencode.global.dat"
if (Test-Path $store) {
Rename-Item -LiteralPath $store -NewName "opencode.global.dat.$stamp.bak"
Write-Host "Backed up GUI store: opencode.global.dat.$stamp.bak"
}
Get-ChildItem -Path $desktopDir -Filter "opencode.global.dat.tmp-*" -ErrorAction SilentlyContinue |
Remove-Item -Force -ErrorAction SilentlyContinue
Create/update OpenCode global config for OpenRouter timeout.
$cfgDir = Join-Path $env:APPDATA "opencode"
$cfg = Join-Path $cfgDir "opencode.json"
New-Item -ItemType Directory -Force -Path $cfgDir | Out-Null
if (Test-Path $cfg) {
Copy-Item -LiteralPath $cfg -Destination "$cfg.$stamp.bak" -Force
$data = Get-Content -LiteralPath $cfg -Raw | ConvertFrom-Json
} else {
$data = [pscustomobject]@{}
}
function Set-Prop($obj, [string]$name, $value) {
if ($obj.PSObject.Properties[$name]) {
$obj.PSObject.Properties[$name].Value = $value
} else {
$obj | Add-Member -NotePropertyName $name -NotePropertyValue $value
}
}
function Ensure-Obj($obj, [string]$name) {
if (-not $obj.PSObject.Properties[$name] -or $null -eq $obj.PSObject.Properties[$name].Value) {
Set-Prop $obj $name ([pscustomobject]@{})
}
return $obj.PSObject.Properties[$name].Value
}
Set-Prop $data '$schema' 'https://opencode.ai/config.json'
$provider = Ensure-Obj $data 'provider'
$openrouter = Ensure-Obj $provider 'openrouter'
$options = Ensure-Obj $openrouter 'options'
Set-Prop $options 'timeout' $false
Set-Prop $options 'chunkTimeout' $chunkTimeoutMs
$data | ConvertTo-Json -Depth 32 | Set-Content -LiteralPath $cfg -Encoding UTF8
Write-Host "Done. OpenRouter chunkTimeout set to $chunkTimeoutMs ms."
Write-Host "Restart OpenCode."
This PS1 script slash chunk should help with the caching issues that makes the GUI feel really laggy especially for me and also the time out
RenanOSX commentedon Jun 16, 2026
Is not an actual fix but the prompt "Do the edits in small batches" does work as you mentioned, thanks!
InfHorus commentedon Jun 21, 2026
I have the same issue, basically OpenRouter is unusable, never seen such unreliable provider, tried 10 differents models they all hit this very same error, I think for a "Router" they should at least make it usable for anything else than sending "hello world" to models.
wasteofoxygen69 commentedon Jun 24, 2026
Same but even on default I get the same error. Also the continue in small baches doesn't really help much it seems.
thimvanamersfoort commentedon Jul 5, 2026
Same issue here with GLM 5.2 on Openrouter. Very frustrating
RenanOSX commentedon Jul 9, 2026
To give more context to this issue, it does not seem to happen on the paid version (GO or Zen plans), only on the free tier models
jimdawdy-hub commentedon Jul 15, 2026
Confirmed on Openrouter using GLM-5.2 (paid) via API plan. It's frequent and should be considered a blocker.
OS: EndeavourOS Linux x86_64
:+sssssssssso/. Host: HP Z640 Workstation-/ossssssssssssso/. Kernel: 7.0.14-zen1-1-zen-/+sssssssssssssssso+:Uptime: 3 days, 22 hours, 7 mins-:/+sssssssssssssssssso+/. Packages: 2524 (pacman).://osssssssssssssssssssso++- Shell: bash 5.3.15.://+ssssssssssssssssssssssso++: Resolution: 1920x1080
.:///ossssssssssssssssssssssssso++: DE: Plasma 6.7.1
:////ssssssssssssssssssssssssssso+++. WM: kwin-////+ssssssssssssssssssssssssssso++++- Theme: Breeze-Dark [GTK2], Breeze [GTK3]..-+oosssssssssssssssssssssssso+++++/Icons: breeze-dark [GTK2/3]./++++++++++++++++++++++++++++++/:. Terminal: konsole
`:::::::::::::::::::::::::------`` CPU: Intel Xeon E5-2699 v4 (44) @ 2.879GHz
GPU: NVIDIA GeForce RTX 5060 Ti
Memory: 36146MiB / 80326MiB
thimvanamersfoort commentedon Jul 15, 2026
I found when using the
xhighreasoning variant (GLM 5.2), this bug seems to happen much less often. Maybe this helps some of you!github-actions commentedon Sep 15, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.