opencode needs to support harmony in order to run gpt-oss-20b #1117

Closed
opened 2026-02-16 17:29:31 -05:00 by yindo · 3 comments
Owner

Originally created by @gtrias on GitHub (Aug 5, 2025).

OpenAI recently released their OSS models: https://github.com/openai/gpt-oss

These models are a promising option for local usage with opencode. However, they are trained to respond using the Harmony format, which opencode currently doesn’t handle correctly.

As a result, responses from gpt-oss are not formatted or parsed properly in opencode:

Example:

Image

I've opened a PR to improve support for the Harmony format: https://github.com/sst/opencode/pull/1632

Originally created by @gtrias on GitHub (Aug 5, 2025). OpenAI recently released their OSS models: https://github.com/openai/gpt-oss These models are a promising option for local usage with opencode. However, they are trained to respond using the [Harmony](https://github.com/openai/harmony) format, which opencode currently doesn’t handle correctly. As a result, responses from gpt-oss are not formatted or parsed properly in opencode: Example: <img width="1770" height="334" alt="Image" src="https://github.com/user-attachments/assets/2fce683d-85f0-4082-8f41-e55f66bad493" /> I've opened a PR to improve support for the Harmony format: https://github.com/sst/opencode/pull/1632
yindo closed this issue 2026-02-16 17:29:31 -05:00
Author
Owner

@x4132 commented on GitHub (Aug 5, 2025):

What inference server are you using? The output you are receiving should be handled at the server level, rather than at the application level.

@x4132 commented on GitHub (Aug 5, 2025): What inference server are you using? The output you are receiving should be handled at the server level, rather than at the application level.
Author
Owner

@gtrias commented on GitHub (Aug 6, 2025):

What inference server are you using? The output you are receiving should be handled at the server level, rather than at the application level.

I'm using lmstudio. It seems the version I was using was not handling the format correctly for some reason. Now it works!

Closing the issue and the PR

@gtrias commented on GitHub (Aug 6, 2025): > What inference server are you using? The output you are receiving should be handled at the server level, rather than at the application level. I'm using lmstudio. It seems the version I was using was not handling the format correctly for some reason. Now it works! Closing the issue and the PR
Author
Owner

@jp127266 commented on GitHub (Aug 8, 2025):

@gtrias, which branch does support the Harmony format?

@jp127266 commented on GitHub (Aug 8, 2025): @gtrias, which branch does support the Harmony format?
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#1117