Repository navigation
[FEATURE]: plugin hook for dynamic model routing based on prompt complexity #18844
Closed as not planned
Description
Feature hasn't been suggested before.
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
Every prompt sent through OpenCode uses the same model, whether it's "hello" or a complex multi-file refactor. There's no way to match model cost to prompt complexity. A developer on Claude ($15/1M tokens) spending 200 simple prompts/day wastes $50-60/month on prompts a free model handles perfectly.
Proposed Solution
Add a chat.model plugin hook that fires before each prompt, allowing plugins to dynamically route messages to different models based on content analysis.
This enables:
- De-escalation: Route simple tasks ("hello", "list files") to free/cheap models
- Escalation (opt-in): Route complex tasks to premium models
- Subagent inheritance: Subtasks inherit the routed model
- Per-prompt override: Users can force their selected model with
!premium
Maybe suggested before but not totally sure:
Found it. There are existing issues that match:
- [FEATURE]: add chat.model plugin hook for pre-call model routing #18793 — "FEATURE: add chat.model plugin hook for pre-call model routing" — OPEN
- [FEATURE]: Plugin can change the active model and have it stick #18667 — "FEATURE: Plugin can change the active model and have it stick" — OPEN
- [FEATURE]: Plugin hook to update active model and have subagents inherit it #18666 — "FEATURE: Plugin hook to update active model and have subagents inherit it" — CLOSED (completed)
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Mar 23, 2026
This issue might be a duplicate of existing issues. Please check:
(Note: you've already identified these yourself in the issue body. Consider closing this issue in favor of one of the above.)
feat: add chat.model hook with reference token-optimizer plugin
JGuck commentedon Jun 3, 2026
I opened #30457 with a minimal implementation of the
chat.modelhook.This should unblock the core plugin-routing part of this request: plugins can now choose a different
{ providerID, modelID }before an LLM call, which makes prompt-complexity routing possible.github-actions commentedon Aug 3, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.