Repository navigation
Compaction triggers too soon in comparison to context window size #21520
Description
Description
Hi,
I'm on github copilot provider and using GPT-5.4 extensively.
From models.dev, GPT-5.4 has:
| Context Limit | Input Limit | Output Limit |
|---|---|---|
| 400,000 | 272,000 | 128,000 |
But opencode triggers compaction early, around 128k.
I can confirm OpenCode is able to read the correct context window:
Also, I added this compaction config but it seems not respected:
{
"compaction": {
"auto": false,
"prune": true,
"reserved": 10000
},
}
Plugins
No response
OpenCode version
1.4.0 TUI
Steps to reproduce
No response
Screenshot and/or share link
Operating System
MacOs
Terminal
Warp
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Apr 8, 2026
This issue might be a duplicate of existing issues. Please check:
compaction.auto: falsebeing ignored when provider throws context overflow — your report of the compaction config not being respected matches this issue closely.github-copilot/gpt-5.4— same provider and model combination experiencing unexpected compaction behavior.If your issue is primarily about the config not being respected, consider commenting on #16882 instead. If it's specifically about the wrong threshold being used for GPT-5.4's context window, this may be a distinct bug worth tracking separately.
github-actions commentedon Jul 3, 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.