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
Thanks for the report! We need a few more details before we can look into this:
OpenCode version: the output of opencode --version, or the Desktop app version.
OS and how you installed OpenCode (npm, brew, the Desktop app, ...).
Provider for Haiku 5.5: Anthropic directly, OpenCode Zen, Bedrock, GitHub Copilot, or another one.
What happens exactly: does /compact get refused, does compaction start and then fail, or does auto-compaction just not start? Please paste the full error message as text. It's also useful to know roughly how full the context was at the time.
Steps: what you were doing when it happened, and whether a retry fixes it.
You can reply here with the details or edit the issue. If you edit it, please also leave a comment saying you've updated it, so we know you're done.
These existing issues may be related, most likely first:
Session compaction failed #54797: compaction fails because the summary request plus the requested output tokens doesn't fit in the model's context window, even though the context limit wasn't hit yet
Thanks for the details, they help. One thing is still missing: I can't read the error in the screenshot, so could you paste it as text? The full message, including any HTTP status code. If you can, please also include:
The exact model ID shown in OpenCode, e.g. opencode-go/... or anthropic/.... That tells us whether this goes through OpenCode Go or straight to Anthropic.
Roughly how many tokens the context showed when you ran /compact.
The relevant lines from the log around the failure, if you can find them (in ~/.local/share/opencode/log/).
These OpenCode Go issues may be related, most likely first. They're about other models and automatic compaction, so please check whether your error looks like either of them:
Session compaction failed #54797: compaction fails when the summary request plus the requested output tokens doesn't fit the model's real context window
Your screenshot shows Haiku 5.5 at max effort running out of output tokens during compaction (32K, which is the cap on the summary). The main problem was that the summary request reused the session's max reasoning setting, so the model spent the whole 32K thinking and never finished the summary. I fixed it by making the summary use the default reasoning instead: #56772
I haven't been able to try it on a real Haiku 5.5 session, although i am sure it will work
Closing this for now. We still need the compaction error as text, and the screenshot couldn't be read on our side. Thanks for the version and platform details.
If it happens again on a current v2 version, please reopen this issue or open a new one with:
the full error message pasted as text
the exact model ID (opencode-go/... or anthropic/...)
roughly how full the context was when you ran /compact
In the meantime, #56296 and #50761 cover similar compaction failures on OpenCode Go and may be worth following.
Activity
opencode-agent commentedon Oct 18, 2026
Thanks for the report! We need a few more details before we can look into this:
opencode --version, or the Desktop app version./compactget refused, does compaction start and then fail, or does auto-compaction just not start? Please paste the full error message as text. It's also useful to know roughly how full the context was at the time.You can reply here with the details or edit the issue. If you edit it, please also leave a comment saying you've updated it, so we know you're done.
These existing issues may be related, most likely first:
suse-coder commentedon Oct 18, 2026
haiku direclty opencode go, ubuntu in wsl, opencode v2.0.32. I just tried to compact it and it gave me that error in the image
opencode-agent commentedon Oct 18, 2026
Thanks for the details, they help. One thing is still missing: I can't read the error in the screenshot, so could you paste it as text? The full message, including any HTTP status code. If you can, please also include:
opencode-go/...oranthropic/.... That tells us whether this goes through OpenCode Go or straight to Anthropic./compact.~/.local/share/opencode/log/).These OpenCode Go issues may be related, most likely first. They're about other models and automatic compaction, so please check whether your error looks like either of them:
This text request is too large for the model's configured input limitbefore OpenCode's context limit is reached, and compaction doesn't recover the sessionGraffian commentedon Oct 18, 2026
Your screenshot shows Haiku 5.5 at max effort running out of output tokens during compaction (32K, which is the cap on the summary). The main problem was that the summary request reused the session's max reasoning setting, so the model spent the whole 32K thinking and never finished the summary. I fixed it by making the summary use the default reasoning instead: #56772
I haven't been able to try it on a real Haiku 5.5 session, although i am sure it will work
opencode-agent commentedon Oct 21, 2026
Closing this for now. We still need the compaction error as text, and the screenshot couldn't be read on our side. Thanks for the version and platform details.
If it happens again on a current v2 version, please reopen this issue or open a new one with:
opencode-go/...oranthropic/...)/compactIn the meantime, #56296 and #50761 cover similar compaction failures on OpenCode Go and may be worth following.
rekram1-node commentedon Oct 21, 2026
i figured out a likely cause btw i have open pr will merge
rekram1-node commentedon Nov 3, 2026
Believe this was addressed 6709f56