[FEATURE]: Zen model disable toggle is not enforced at the gateway - Codex keeps auto-using gpt-5.6-luna and billing continues #41697

Activity

1. github-actions

   assigned

   fwang

   on Aug 11, 2026
2. github-actions commented on Aug 11, 2026

   

   github-actionsBot

   on Aug 11, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue doesn't fully meet our contributing guidelines.

   What needs to be fixed:

   • Feature requests must use the [FEATURE]: template: the title should start with [FEATURE]: and the issue body should include the required verification checkbox confirming you've searched for existing issues.

   Please edit this issue to address the above within 2 hours, or it will be automatically closed.

   ---

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

   • [FEATURE/BUG] OpenCode Zen disabled models still appear in TUI picker (config drift risk) #40560: Zen disabled models still appear in the TUI model picker and are not filtered by the gateway (same core problem of model-disable not being enforced server-side)

   If you believe this was flagged incorrectly, please let a maintainer know.
3. github-actions

   added

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   on Aug 11, 2026
4. YiRan0

   changed the title [-]Feature request: Zen model disable toggle is not enforced at the gateway - Codex keeps auto-using gpt-5.6-luna and billing continues[/-] [+][FEATURE]: Zen model disable toggle is not enforced at the gateway - Codex keeps auto-using gpt-5.6-luna and billing continues[/+] on Aug 11, 2026
5. YiRan0 commented on Aug 11, 2026

   

   YiRan0

   on Aug 11, 2026

   Author

   More actions

   Regarding the duplicate flag for #40560: this is related but not the same issue.

   • [FEATURE/BUG] OpenCode Zen disabled models still appear in TUI picker (config drift risk) #40560 = TUI model picker / /models catalog still shows disabled models (display/UX), and selecting one then fails at the gateway.
   • This issue = the gateway does not reliably reject or stop billing for a disabled model. After disabling gpt-5.6-luna in Zen, a direct POST /responses with that model still returns HTTP 200, completes, and generates billing records. The caller is Codex's internal helper agent, which hardcodes the model slug and cannot be configured client-side.

   So even a complete fix for #40560 (catalog filtering + picker UX) would not stop the cost in this scenario; only server-side enforcement (reject disabled model + no billing, per-key allowlist/blocklist, spend cap) can. I've also added the [FEATURE]: prefix and the verification checkbox to comply with the contributing guidelines.
6. github-actions

   removed

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   on Aug 11, 2026
7. github-actions commented on Aug 11, 2026

   

   github-actionsBot

   on Aug 11, 2026 – with GitHub Actions

   Contributor

   More actions

   Thank you for updating your issue! It now clearly explains the problem, includes steps to reproduce, environment details, and references related issues. Everything looks good — we appreciate the thorough write-up!
8. sergiopadure commented on Oct 3, 2026

   

   sergiopadure

   on Oct 3, 2026

   Contributor

   More actions

   Same problem on an OpenCode Go subscription, and it goes beyond individual models: disabling the whole provider has no effect either.

   On 2026-10-03 I disabled the models I don't use (including qwen3.8-flash), then every model in the OpenCode Go model list, and finally the OpenCode Go provider entirely. The same Go API key keeps working throughout:

   • OpenCode v2.0.22, headless: opencode run -m opencode-go/qwen3.8-flash "Reply with exactly: OK" exits 0 with an answer from qwen3.8-flash. I ran it three times between 16:40 and 16:44 UTC, after qwen3.8-flash was disabled.
   • A separate OpenAI-compatible client, same key: chat completions for qwen3.8-flash returned HTTP 200 with tokens consumed, at 16:35 and 16:43 UTC.
   • After disabling the provider entirely: my own test later that afternoon still got answers.

   Expected: requests for a disabled model, and every request once the provider is disabled, are rejected (for example HTTP 403) and are not billed.

   As it stands, a subscriber has no way to stop a client from using, and being billed for, a model they have switched off, so the toggles give a false sense of control. This looks like a gateway enforcement bug rather than a feature request.

   Could someone from the team confirm whether these toggles are meant to be enforced at the gateway, and if so, prioritise the fix?
9. gustavo-oo commented on Oct 6, 2026

   

   gustavo-oo

   on Oct 6, 2026

   Last edited by gustavo-oo

   More actions

   Adding another confirmation on an OpenCode Go subscription — this time the caller was OpenCode's own subagent, and the model has a small monthly allowance.

   Scenario

   • Workspace: wrk_01KS5JJ92HP249G45Y7YVT0GPF
   • I disabled Kimi K3 in the workspace Models tab.
   • Usage by model still shows Kimi K3 = $1.65 in the last 7 days. The calls came from one of my own OpenCode subagents (the task tool), not from an external client. Screenshots of the usage page and the disabled toggle are attached.

   Direct repro against the Go gateway (today, after the model was disabled)

   curl https://opencode.ai/zen/go/v1/chat/completions \
     -H "Authorization: Bearer <Go key>" \
     -H "x-opencode-session: ses_issue-repro-kimi-disabled" \
     -H "User-Agent: opencode/2.0.20" \
     -H "Content-Type: application/json" \
     -d '{"model":"kimi-k3","messages":[{"role":"user","content":"Reply with the single word: pong"}],"max_tokens":8}'

   Result: HTTP 200 with tokens billed:

   {"id":"727e4cf4d53bc3ec819ed8979335dcf9","object":"chat.completion","created":1791305279,"model":"kimi-k3","choices":[{"index":0,"finish_reason":"length","message":{"role":"assistant","content":"","reasoning_content":"The user wants me to reply with just"}}],"usage":{"prompt_tokens":92,"completion_tokens":8,"total_tokens":100,"prompt_tokens_details":{"audio_tokens":0,"cached_tokens":0,"cache_creation_input_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{}}}

   Impact

   Kimi K3 has a $15/month allowance on Go, so $1.65 is ~11% of that model's allowance spent on a model I had switched off. As the comment above notes, the toggle gives a false sense of control.

   Expected

   The gateway should reject requests for workspace-disabled models (e.g. HTTP 403 with a model_disabled error) and not bill them, regardless of which client or subagent sends the request.

   Environment

   • opencode v2.0.20
   • macOS 25.6.0 (arm64), iTerm2, zsh
   • Subscription: OpenCode Go

   
10. opencodejr commented on Oct 28, 2026

    

    opencodejrBot

    on Oct 28, 2026

    Contributor

    More actions

    Thanks for the detailed write-up, and for tracking down the Codex side of this.

    To restate: you'd like the OpenCode Zen/Go gateway to enforce the Models tab toggle on the server. Requests for a model you've turned off would be rejected and not billed. You'd also like per-key model allow/block lists and spend/token caps as a safety net, because clients like Codex hardcode their helper-agent models and can't be configured to avoid them.

    This can't be done in the OpenCode app or with a plugin, because the requests come from Codex and only the gateway can block them. Maintainers will consider it, though there's no commitment on whether or when.

    Related issues:

    • console: workspace Models-tab enablement not honored - disabled models still selectable and billed #54929: same problem with a Console workspace (turned-off models still used and billed), closed as a duplicate of this one
    • [FEATURE/BUG] OpenCode Zen disabled models still appear in TUI picker (config drift risk) #40560: turned-off models still shown in the OpenCode TUI model picker
    • [Bug] OpenCode Go bills gpt-5.6-terra requests as gpt-5.6-luna (Codex Memories background jobs) #42342: Codex background jobs on Go billed as gpt-5.6-luna
    • [FEATURE]: Suggestion: a small fail-safe for Zen keys — pause LLM usage over the wire, reversibly #52066: request for a fail-safe to pause usage on a Zen key

    If you'd like the gpt-5.6-luna charges reviewed, billing isn't handled in GitHub issues. Please email [email protected] with the details, and keep account and payment information out of this public issue.