[Bug] Leading space in model name when using opencode-go/deepseek-v4-flash

[lcobos-dev]

Description

Error from provider (Console Go): Upstream request failed: [invalid_request_error]
The supported API model names are deepseek-v4-pro or deepseek-v4-flash,
but you passed deepseek-v4-flash.

• OpenCode Desktop v1.18.15 (Windows 11)
• OpenCode Go subscription
• /model opencode-go/deepseek-v4-flash fails
• /model opencode-go/deepseek-v4-pro works fine
• Config file is correct (verified with hex dump)
• Cleared cache, workspace state, SharedStorage — error persists

Plugins

No response

OpenCode version

OpenCode Desktop v1.18.15 (Windows 11)

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[1123Chase1123]

The same problem！

[Slacom]

Same issue here on OpenCode Go. deepseek-v4-flash fails, deepseek-v4-pro works. Reproducible with a clean curl request, so it's not a local config problem. Looks like a server-side issue with the Console Go gateway.

[Slacom]

Btw it still runs fine in the OpenCode TUI

[chaosmanage]

Confirming this is an issue for us too. OpenCode Go + deepseek-v4-flash fails with the identical error — ...but you passed deepseek-v4-flash (note the double space) — while deepseek-v4-pro and other Go models return HTTP 200 on the same key/endpoint. We hit it both in the TUI and from sub-agents, and it's reproducible with a direct request to https://opencode.ai/zen/go/v1/chat/completions using the documented model id. Please prioritize a fix.

[Skylvw]

Diagnosis: this is a server-side gateway bug, not a client bug — nothing in this repo can fix it.

Evidence:

• GET https://opencode.ai/zen/go/v1/models (no auth needed) returns a clean list: 25 models, no ids with whitespace, no duplicates, deepseek-v4-flash present exactly once.
• Yet a plain curl to /zen/go/v1/chat/completions with "model":"deepseek-v4-flash" gets rejected with ...you passed deepseek-v4-flash (note the double space = leading space) while deepseek-v4-pro works on the same key/endpoint.

So the advertised model list is fine, but the gateway's per-model upstream routing entry for deepseek-v4-flash is mangling the id (leading space in the mapping table, likely in the upstream id or a route key). Since the error comes straight from the upstream provider echoing back the mangled name, no client-side workaround can help — the gateway must trim/normalize the model id before forwarding (or the routing entry needs fixing).

Flagging for the opencode.ai gateway maintainers. There is no in-repo fix; the opencode client forwards the id exactly as documented.

[fwang]

should be fixed now