SSE error with #44944 — broken

[Artyomkun]

Description

The Big Pickle model was observed entering a loop.

Plugins

No response

OpenCode version

1.18.27 ( Also 1.18.26)

Steps to reproduce

You just generate the code, and at the moment the AI ​​pauses to think, it breaks out of the loop—though sometimes the edit gets messed up.

Screenshot and/or share link

No response

Operating System

Windows 11

Terminal

PowerShell

[github-actions]

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

• big-pickle got stuck in an autoregressive loop during a brainstorming task #42956: big-pickle got stuck in an autoregressive loop during a brainstorming task (same model, same loop symptom)
• Bug: OpenCode enters infinite loop after tool calls complete (Zen/big-pickle) #26220: OpenCode enters infinite loop after tool calls complete (Zen/big-pickle, same model looping behavior)

Also note: issue #44944 referenced in the title does not appear to exist in this repository — you may want to double-check the issue number you intended to reference.

[Artyomkun]

#42956: big-pickle got stuck in an autoregressive loop during a brainstorming task (same model, same loop symptom) The loop is still there, and it keeps speaking mostly in English. At first, I ask it to speak in my native language—Russian—but then it immediately switches back to English. Version 1.18.25.

#26220: OpenCode enters infinite loop after tool calls complete (Zen/big-pickle, same model looping behavior)
That issue doesn't look like the one I had—back then, the model was writing away when OpenCode suddenly stopped it. I noticed this yesterday; the version was 1.18.25. Over the course of the day, it updated to version 26 and then 27. The same thing happened on 27; I didn't see that bug on 26—it might have been there, but I didn't come across it.

@AlexanderWillner:
fix(provider): handle SSE reader cancel rejections (#44944) (relaese 1.18.27)

[Artyomkun]

Another thing I noticed is that OpenCode stops abruptly when a message isn't fully generated, and I lose time because of this.

[Artyomkun]

I noticed a difference between two project chats: one is fine, but in the other, the AI ​​cuts off mid-sentence and I see <reasoning> context </reasoning>. It looks like an escaping bug in the stream.

[Artyomkun]

To sum up, the interruption bug can manifest in various ways. I suspect the issue lies somewhere within the branching logic that handles the message response threads.

[Artyomkun]

A second hypothesis is that the essay stream sometimes breaks during editing; this might be linked to the halting of the essay message.

[Artyomkun]

I’ve noticed something else now that makes it clearer: the model itself doesn't stop—the stream stops when it decides to change something in a file or execute a command. Only then does OpenCode think the response is finished and halt the stream, even though the message context isn't actually over; the message just gets cut off (it only holds the message for a maximum of 3 minutes).

[Artyomkun]

When writing commands or updating the to-do list, the escaping breaks during the stream, causing the chat to stop receiving new messages.

[Artyomkun]

But this is noticeable: if a chat had to download three patches at once, it breaks for some reason, whereas new chats work fine. That’s a new observation.