OpenCode Go quota is being consumed by models not present in my OpenCode/OMO automatic routing

Description

Problem

I'm seeing unexpectedly high OpenCode Go quota usage from models that I am not intentionally using and that are not present in my current OpenCode/OMO automatic routing.

The OpenCode Go Usage dashboard shows usage from models including:

• DeepSeek V4 Flash
• DeepSeek V4 Pro
• GLM-5.2
• GLM-5.3
• Kimi K3

DeepSeek V4 Flash and GLM-5.2 in particular have consumed a significant amount of quota.

Some individual usage records show very large input contexts, around 80K–100K input tokens, which causes my Go quota to decrease very quickly.

I have already inspected and cleaned up my local OpenCode and OMO automatic routing, but these unexpected model usage records still appear.

---

Environment

• OS: Windows 11
• OpenCode: 1.18.21
• OMO (oh-my-openagent): 4.19.4
• OpenCode Go subscription: Active

I use OpenCode together with OMO for agent routing and delegation.

---

Current OpenCode configuration

My current main OpenCode models are:

• default: openai/gpt-5.6-luna
• small_model: opencode/x-preview-f-free

The previous DeepSeek Free model has already been removed from my small_model configuration.

---

Current OMO routing

My current OMO routing includes:

• Sisyphus: openai/gpt-5.6-luna
• Explore: opencode/x-preview-f-free
• Librarian: opencode/x-preview-f-free
• Sisyphus-Junior: opencode/x-preview-f-free
• deep: opencode/x-preview-f-free
• quick: opencode/x-preview-f-free
• Hephaestus: opencodex/cursor/composer-2.5

I inspected both the OMO configuration and the resolved routing after restarting OpenCode.

For high-frequency and lower-cost tasks, I am currently intentionally using Ox Alpha through:

opencode/x-preview-f-free

---

Current automatic OpenCode Go routes

I have already removed unwanted Go models from my automatic routing.

The only OpenCode Go models intentionally remaining in automatic routes are:

• opencode-go/ox-alpha-free
• opencode-go/mimo-v2.5
• opencode-go/muse-spark-1.2-contributor

The following models are NOT present in my current OpenCode/OMO automatic routing:

• opencode-go/deepseek-v4-flash
• opencode-go/deepseek-v4-pro
• opencode-go/glm-5.2
• opencode-go/glm-5.3
• opencode-go/kimi-k3
• opencode-go/qwen3.7-max
• opencode-go/qwen3.8-max

These models may still exist in the provider model catalog and may be manually selectable, but they are not configured as automatic routes.

I am not claiming that they have been removed from the OpenCode model catalog. I am specifically referring to the active automatic routing configuration.

---

Reproduction / observation

After cleaning up the routing configuration and restarting OpenCode, I ran a very small OMO configuration health check.

The test was intentionally limited to a single Explore delegation.

The local delegation completed successfully and showed:

opencode/x-preview-f-free

My local configuration inspection also confirmed:

• Explore: opencode/x-preview-f-free
• Librarian: opencode/x-preview-f-free
• Sisyphus-Junior: opencode/x-preview-f-free
• deep: opencode/x-preview-f-free
• quick: opencode/x-preview-f-free

The automatic Go routes were also confirmed to contain only:

• opencode-go/ox-alpha-free
• opencode-go/mimo-v2.5
• opencode-go/muse-spark-1.2-contributor

However, during/around this small test period, the OpenCode Go Usage dashboard showed new usage from models including:

• DeepSeek V4 Flash
• DeepSeek V4 Pro
• GLM-5.2
• Kimi K3

Some examples in the Usage dashboard showed approximately:

• DeepSeek V4 Flash: ~80K–100K input tokens on individual requests
• GLM-5.2: ~83K input tokens on an individual request
• DeepSeek V4 Pro: ~40K–50K input tokens on individual requests
• Kimi K3: additional usage also appeared

This caused my OpenCode Go quota to decrease much faster than expected.

---

Local investigation

I also checked my local OpenCode configuration, OMO configuration, resolved routing, OpenCode database, and available logs.

I cannot find DeepSeek V4 Pro, GLM-5.2, GLM-5.3, or Kimi K3 in the current automatic routing configuration.

For the small test described above, the local delegation indicated:

opencode/x-preview-f-free

I have not been able to correlate the unexpected DeepSeek / GLM / Kimi usage records shown in the Go Usage dashboard with corresponding local requests to those models.

Therefore, I currently do not know what is generating these Go usage records.

I do not want to speculate about whether this is caused by OpenCode, OMO, fallback behavior, usage reporting, or something else.

I would like clarification from the OpenCode team based on the account-side/server-side usage information.

---

Expected behavior

If my active automatic routing does not contain DeepSeek V4 Pro, GLM-5.2, GLM-5.3, Kimi K3, etc., I would not expect these models to consume significant amounts of my OpenCode Go quota unless I explicitly select or configure them.

At minimum, I would like to be able to determine which local request/session/client generated each Go Usage record.

---

Request for clarification

Could the OpenCode team please help clarify why these Go usage records are appearing on my account?

Specifically:

1. What is generating the DeepSeek V4 Flash, DeepSeek V4 Pro, GLM-5.2, GLM-5.3, and Kimi K3 usage shown in my OpenCode Go account?

2. Why do these usage records continue to appear even though these models are not present in my current OpenCode/OMO automatic routing?

3. Is there a way to correlate each OpenCode Go Usage record with its originating:

   • OpenCode session
   • request
   • agent/subagent
   • client
   • requested model

4. Is there a request ID or other identifier available in the Go Usage dashboard/API that can be matched against local OpenCode logs?

5. Is there a way to restrict OpenCode Go automatic usage to an explicit allowlist of models?

My intended automatic Go models are currently only:

• Ox Alpha Free
• MiMo V2.5
• Muse Spark 1.2 Contributor

The unexpected DeepSeek / GLM / Kimi usage is consuming my Go quota very quickly, especially when individual requests contain 80K–100K input tokens.

I would appreciate clarification on where these requests are coming from and how I can prevent models outside my intended automatic routing from consuming my Go quota.

I can provide screenshots of the Go Usage dashboard, my OpenCode configuration, OMO routing configuration, and additional local telemetry/logs if needed.

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response