Same model cannot use different reasoningEffort values across agents

Description

When two agents use the same model with different reasoningEffort values, OpenCode appears to collapse them into a single variant.

Example:

{
"agent": {
"plan": {
"model": "deepinfra/deepseek-ai/DeepSeek-V4-Flash",
"reasoningEffort": "xhigh"
},
"build": {
"model": "deepinfra/deepseek-ai/DeepSeek-V4-Flash",
"reasoningEffort": "medium"
}
}
}

Expected

PLAN uses xhigh
BUILD uses medium

Actual

Both agents end up using the same reasoning variant (the /variants UI also shows the same value for both).

If both agents are configured with the same value (xhigh/xhigh or medium/medium), everything works correctly.

Plugins

No response

OpenCode version

1.18.7

Steps to reproduce

1. Use the provided opencode.json
2. Observe that the UI does not maintain a different "reasoningEffort" for PLAN and BUILD while toggling between modes.

IMPORTANT : This issue concerns only config that use same model but differents reasoningEffort for PLAN and BUILD

Screenshot and/or share link

Operating System

Windows 10

Terminal

Windows terminal / Arch WSL