Repository navigation
OpenCode Go: DeepSeek V4.1 Flash returns HTTP 500 while GLM works with the same credentials #56721
Description
Description
- Windows x64
- OpenCode CLI v2.0.32
- Provider: opencode-go
- Model: deepseek-v4.1-flash
- Variant: low
- Date: 2026-10-18, Asia/Taipei
Problem:
DeepSeek V4.1 Flash consistently fails with:
HTTP 500
inference_failed: The inference request failed.
Endpoint:
POST /inference/go/openai/v1/chat/completions
Request ID from Console:
01a14e0a-fe95-79f9-a521-6890d76c24da
Minimal reproduction:
opencode run --agent plan
--model opencode-go/deepseek-v4.1-flash#low
"Do not use tools. Reply OK only."
Tests performed:
- Standalone and background-service modes both fail.
- A minimal direct API request also returns 500 with an empty body.
- Omitting reasoning_effort does not help.
- The legacy /zen/go/v1/chat/completions endpoint also fails.
- Go usage reports rolling, weekly and monthly status as OK.
- GLM 5.3 Flash succeeds through both API and CLI using the same credentials.
Please investigate the DeepSeek-specific inference route.
The exact upstream cause is unknown.
Plugins
none
OpenCode version
OpenCode CLI v2.0.32
Steps to reproduce
No response
Screenshot and/or share link
Operating System
window11
Terminal
window terminal
Activity
potato-fried-potato commentedon Oct 21, 2026
I hit the same DeepSeek-specific failure on 2026-10-21.
deepseek-v4.1-flashPOST /inference/go/openai/v1/chat/completions0.163.0-alpha.10.22026-10-21 10:54:05 +08:00429,inference_failed, "The inference request failed."The Go usage dashboard shows my rolling, weekly, and monthly usage as available. Other OpenCode Go models work with the same provider and credentials, so this looks DeepSeek/upstream-specific rather than a client or subscription issue.
neriousy commentedon Oct 21, 2026
Thanks for the report, and sorry for the disruption. This was a temporary problem with the hosted service or an upstream model provider, and it has since recovered. You can see live status and past incidents at https://status.opencode.ai. If you still see this now, please open a new issue with the time (with timezone), model, and request ID from the console logs.
elitesmitpatel commentedon Oct 24, 2026