Subagent fallback chain prepends incorrect opencode/ prefix to model names, causing infinite retry loop

[maxxua]

Description

Description

When a background subagent task is spawned (e.g. task(category="unspecified-high", run_in_background=true)), the initially assigned model nvidia/z-ai/glm-5.1 returns HTTP 429 (Too Many Requests). The system then enters a fallback chain, but every fallback model name incorrectly includes an opencode/ prefix that the platform's model registry does not recognize.

Error Chain

Initial: nvidia/z-ai/glm-5.1 → 429
Fallback #1: opencode/claude-opus-4-7 → Model not found. Did you mean: claude-opus-4-7?
Fallback #2: opencode/gpt-5.5 → Model not found. Did you mean: gpt-5.5, gpt-5.5-pro?
Fallback #3: opencode/glm-5 → Model not found. Did you mean: glm-5, glm-5.2, glm-5.1?
Fallback #4: opencode/kimi-k2.5 → (same)

Root Cause

Two issues:

1. Default subagent model nvidia/z-ai/glm-5.1 consistently returns 429 — should be replaced with an available model.
2. Fallback chain prepends opencode/ to model names (e.g. opencode/claude-opus-4-7), but the platform expects bare names like claude-opus-4-7. The error messages confirm this with "Did you mean: X".

Impact

• Background subagents (explore, librarian, oracle, etc.) fail unrecoverably when primary model is unavailable
• Wastes significant tokens on retry loops

Environment

• User model: opencode/deepseek-v4-flash-free (main session works fine)
• Plugin: @opencode-ai/plugin@1.14.18
• Platform: opencode.ai SaaS
• OS: WSL2 Ubuntu

Plugins

omo, superpowers, openspec

OpenCode version

OpenCode 1.17.9

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[github-actions]

This issue might be a duplicate of or closely related to an existing issue. Please check:

• Zen gateway loses provider retry timing and retries 429s early #33955: Zen gateway loses provider retry timing and retries 429s early — also involves 429 handling and retry behavior in the Zen/Console gateway. The opencode/ prefix prepended to fallback model names you describe may be a related or downstream symptom of the same gateway-side retry/routing logic.

[rekram1-node]

we dont have a fallback chain, this must be some misunderstsanding