Qwen 3.8 Max weekly limit blocks all other Go models, including unused models #51550

Description

I subscribed to OpenCode Go on September 25, 2026, and started using it the same day.
Most of my usage was on qwen3.8-max. According to my exported request logs, my total successful Qwen 3.8 Max usage was approximately $7.46.
My Go dashboard currently shows:

• Rolling usage: 0%
• Weekly usage: 100%
• Monthly usage: 50%
  I understand that Qwen 3.8 Max has a lower effective quota and that approximately $7.50 of Qwen usage can correspond to the weekly limit.
  However, after reaching the Qwen 3.8 Max limit, I can no longer use other Go models that I have not used at all.
  For example:
• kimi-k2.6 → rejected with HTTP 429
• qwen3.7-plus → rejected with HTTP 429
• gpt-5.6-luna → rejected with HTTP 429
• kimi-k2.7-code → rejected with HTTP 402 / inference_failed
  According to my request log export, I have no successful paid usage on the Kimi models. The Kimi requests were rejected before any tokens were generated or billed.
  This means that reaching the Qwen 3.8 Max weekly quota appears to block the entire Go model pool, instead of only blocking the model whose individual quota was exhausted.
  Expected behavior
  After reaching the weekly quota for Qwen 3.8 Max, I would expect Qwen 3.8 Max to become unavailable, while models with remaining individual quota, such as Kimi, should still be usable.
  Actual behavior
  Once the Qwen 3.8 Max quota was reached, all other Go models also started returning 429 or 402 errors, even though they had no successful usage.
  Free models still work.
  Additional information
• OpenCode version: 1.18.32
• Product: go
• Client: OpenCode CLI
• Subscription started: September 25, 2026
• Issue reproduced immediately after the Qwen weekly quota was reached
  I will attach:

1. My exported request logs CSV
2. A screenshot of the Go usage dashboard showing Weekly 100% and Monthly 50%
   Could you please confirm whether Go quotas are supposed to be shared globally across all models, or whether this is a quota enforcement bug?
   If quotas are model-specific as documented, then the current behavior seems incorrect because unused models are being blocked as soon as one model reaches its limit.

Here is my all usage csv
https://www.transfernow.net/dl/20260926oQ1AAlxV

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response