Repository navigation
API inference blocked while authenticated model catalog remains accessible #46219
Closed as not planned
Description
Description
My OpenCode account has a positive balance, but inference is consistently rejected with HTTP 401 and the error code INFERENCE_ACCESS_BLOCKED.
The same API keys can fetch both model catalogs successfully (HTTP 200), so the credentials are valid enough for authenticated catalog access. Minimal inference requests fail for both:
- OpenCode Go:
deepseek-v4-flashvia/zen/go/v1/chat/completions - OpenCode Zen:
claude-sonnet-5via/zen/v1/messages
I reproduced this with several newly generated API keys on the same account, including on 2026-08-30 UTC. These were isolated one-turn plain-text requests with no tools, files, multimedia, large context, parallelism, or retry loop.
Could you please clarify:
- Is the account or workspace blocked from inference?
- What policy or event caused the block?
- Is it temporary, and what is the process to appeal or restore access?
- Is there a private channel where I can provide the account identifier and exact request timestamps without publishing credentials?
No API keys are included in this report.
Plugins
None; reproduced against the documented HTTP endpoints directly.
OpenCode version
Not applicable; direct API request.
Steps to reproduce
- Create a new API key from an account with positive balance.
- Fetch the Go or Zen
/modelsendpoint: HTTP 200. - Send a minimal single-user-message inference request to one of the documented model endpoints.
- Observe HTTP 401 with
INFERENCE_ACCESS_BLOCKED.
Operating System
Linux
Terminal
Direct HTTPS client
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Aug 30, 2026
This issue might be a duplicate of existing issues. Please check:
chat/completionsreturns HTTP 401 while/v1/modelssucceeds with the same key, affecting OpenCode Go/Zen endpointsRequest blocked by upstream provider(HTTP 401) despite active subscription and valid API keyIf your situation matches one of these, consider adding a comment there rather than opening a new issue.
OdinManiac commentedon Aug 30, 2026
Additional reproduction on 2026-08-30 UTC: another newly generated API key on the same account shows the same split behavior: authenticated model catalog requests succeed, while minimal one-turn inference is rejected with HTTP 401 /
INFERENCE_ACCESS_BLOCKED. No API key or OAuth credential is included here. This still looks account/workspace-level rather than key-level, model-level, balance-related, or caused by request volume.OdinManiac commentedon Aug 30, 2026
A stronger diagnostic signal from a fresh key on the same account (2026-08-30 UTC):
/models: HTTP 200/models: HTTP 200deepseek-v4-flash: HTTP 401AuthErrorclaude-sonnet-5: HTTP 401AuthErrorBoth inference responses explicitly state that the account was blocked for suspected fraud or a terms-of-service breach and direct the user to
help@anoma.ly. This confirms that the failure is account-level, not caused by an invalid key, insufficient balance, a missing model, request size, tools, multimedia, concurrency, or retry volume.Could a maintainer confirm whether this GitHub issue is sufficient for investigation, or whether the account identifier and request timestamps must be sent only by email? I can provide them privately. No API key or OAuth credential is included here.
neriousy commentedon Oct 20, 2026
Thanks for the careful diagnosis, and for keeping credentials out of the report. You're right that this is an account-level block rather than a key, model or balance problem. Blocks like this are reviewed manually, and there's no self-serve way to lift them, so the private channel you asked about is help@anoma.ly. Include your account email, workspace name and a couple of request timestamps, and the team can tell you why it happened and whether access can be restored. Closing here since GitHub can't see account data.