When using OpenCode with 9Router, the percentage of context windows being used always appears as 0%.

[void0x14]

Description

I'm using OpenCode with 9Router, and it's not showing the percentage of the context window that's being used.

Normally, it shows up when I select and use a different model from the models section, but it doesn't show it for anything used with 9Router.

Plugins

@plannotator/opencode @targutnen/opencode-smart-title opencode-agent-tmux opencode-anttgravity-auth opencode-ktlo-auth opencode-mem opencode-mystatus opencode-gwencode-oauth opencode-windsurf-auth opentmux superpowers

OpenCode version

1.14.25

Steps to reproduce

Create a combo in 9router, then select OpenCode from the CLI Tools tab, go to the Models section, and click the Add Models button. Next, select the combo you created from the Combos section, or search for it to select it. Click the “Apply” button, then open OpenCode and enter any text into the model (since OpenCode opens by default with the combo you selected, you don’t need to manually select the model from the “Models” section; however, if it doesn’t appear, you can manually search for and select the combo you created from the ‘Models’ section), then monitor the “%0 used” section.

Screenshot and/or share link

Operating System

Cachyos 7.0.1-1-cachyos-bore-lto

Terminal

Konsole

[github-actions]

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

• Token usage always shows 0 when using @ai-sdk/openai with custom baseURL (Responses API) #20885: Token usage always shows 0 when using @ai-sdk/openai with a custom baseURL/proxy — same symptom of 0% used when routing through a custom provider/proxy

[FCCMac]

experiencing the same issue with LM Studio as well

[void0x14]

experiencing the same issue with LM Studio as well

Man, honestly, don’t even bother waiting—they’ll never come up with a solution for this, and they probably won’t accept the PR either. So you’ll either have to stick with your own fork or just let it go. I went with the second option, to be honest.

[rekram1-node]

The issue is simply opencode doesn't know the context size of your model, so if you share more about your setup (such as: opencode debug config output) I can help

Also I do merge prs and need to close old ones but the vast vast majority are 100% slop

[void0x14]

The issue is simply opencode doesn't know the context size of your model, so if you share more about your setup (such as: opencode debug config output) I can help

Also I do merge prs and need to close old ones but the vast vast majority are 100% slop

So, if we’re talking about prioritization, yes, I agree it’s unnecessary, but it’s still one of the features that should be included in the final release. Also, I just couldn’t get into this OpenCode debug thing—it feels like it’s really lacking in user experience and it’s been a real pain, so unfortunately I can’t share it right now, but I’ll contribute to the code once I’m done with my current work. We love using it, even though we complain about it—we just can’t seem to let go of OpenCode :D.

[rekram1-node]

If youre using a provider thats not included out of the box the ux is significantly degraded, the vast majority of users dont have that issue since there are like over 100 providers out of the box but I dont think 9Router is included in that. So yeah big gap for us currently.

[FCCMac]

experiencing the same issue with LM Studio as well

So the docs don't specify this anywhere that I could easily find it, but yes for some provider/model combinations it seems you have to specify the context limit in the provider block:

"qwen/qwen3.6-35b-a3b": {
    "name": "Qwen3.6 35B a3b (local)",
    "options": {
        ...
    },
    "limit": {
        "context": 171352,
        "output": 20000
    }
}

this has me patched up. I could have swore that I didn't declare this with another local model, but maybe I'm mistaken. thanks all and hopefully this works for you as well @void0x14 !

[void0x14]

Ben de LM Studio ile aynı sorunu yaşıyorum.

Dokümanlarda bunu kolayca bulabileceğim bir yerde belirtilmiyor, ancak evet, bazı sağlayıcı/model kombinasyonları için bağlam sınırını sağlayıcı bloğunda belirtmeniz gerekiyor gibi görünüyor:

"qwen/qwen3.6-35b-a3b": {
"name": "Qwen3.6 35B a3b (local)",
"options": {
...
},
"limit": {
"context": 171352,
"output": 20000
}
}
Bu sayede sorunum çözüldü. Bunu başka bir yerel modelle tanımlamadığımdan emindim, ama belki de yanılıyorum. Herkese teşekkürler ve umarım bu sizin için de işe yarar.@void0x14!

Thank you; I’ll give this a try for a GPT 5.5 score as soon as possible.