Go plan: free-model congestion + silent quota exhaustion stalls all work with no warning (Windows) #41684

Description

I am an opencode Go subscriber (Taiwan, GMT+8) and I want to report two problems that together caused me to burn $12 of quota in 45 minutes with zero output, and I had no idea my quota was exhausted until everything silently stopped.

Problem 1 ??Free model (big-pickle) congestion is silent and unresponsive for hours.
On 2026-08-10 (UTC) my big-pickle sessions kept failing with Rate limit exceeded. Please try again later. (16+ occurrences between 11:18 and 15:39, confirmed in the local log %LOCALAPPDATA%\.local\share\opencode\log\opencode.log). After 15:39 the model went completely silent for ~4 hours (next record only at 19:24). During that time I got no responses and no error surfaced in the UI ??I only saw a stuck session. I assumed it was general network congestion.

Problem 2 ??When I switched to paid models to unblock myself, quota drained with no warning, and then everything stopped silently.
At 19:26 I switched to the Go paid models (deepseek-v4-pro, then glm-5.1) because big-pickle was unresponsive. Over ~45 minutes I made 254 paid calls (glm-5.1 ?158, deepseek-v4-pro ?95, deepseek-v4-flash ?1). At 20:07 the log shows AI_APICallError: Insufficient balance. Manage your billing here: .../billing and every session stopped mid-task. Weekly usage jumped to 40% in that single burst. I never saw an in-app quota warning; the first I knew was that my tasks silently stopped with no completed results.

The result: I paid for a plan, my work stalled for hours due to the free-tier congestion, the workaround (switching to paid models) consumed a large chunk of quota in under an hour, and I finished with no output and no clear error explaining why. My goal here is not a refund ??I want the congestion and the lack of warning/visibility fixed.

Plugins

None (stock opencode Desktop)

OpenCode version

OpenCode Desktop 1.18.10

Steps to reproduce

1. Use the free big-pickle model during a period of congestion ??observe Rate limit exceeded. Please try again later. errors in the local log, followed by several hours with no responses and no surfaced error in the UI.
2. Switch to a Go paid model (e.g. deepseek-v4-pro / glm-5.1) to unblock the stuck session.
3. Run agentic work with parallel subagents ??observe the rolling 5-hour quota ($12) being exhausted within ~45 minutes.
4. When the quota is exhausted, the log records Insufficient balance but there is no in-app warning beforehand, and sessions stop mid-task with no completed results.

Screenshot and/or share link

Local log evidence (paths): C:\Users\hpspy\.local\share\opencode\log\opencode.log

• 16+ Rate limit exceeded errors for big-pickle between 2026-08-10T11:18 and 15:39 UTC
• big-pickle silent 2026-08-10T15:39 ??19:24 UTC (~4h)
• 254 paid-model stream calls 2026-08-10T19:26 ??20:11 UTC
• Insufficient balance errors from 2026-08-10T20:07 UTC

Operating System

Windows 11 (Asia/Taipei, GMT+8)

Terminal

OpenCode Desktop (1.18.10), built-in terminal