[BUG] GLM Coding Plan Pro ($15/mo) unusable due to undocumented concurrent request limit of 1 #8618

Environment

• OpenCode Version: 1.1.x
• OS: macOS Sequoia (Darwin 25.2.0)
• Provider: zai-coding-plan (Coding Plan Pro - $15/mo)
• Model: GLM 4.7

Description

I'm a paying subscriber to Z.ai's Coding Plan Pro tier ($15/month, 600 prompts/5hr window). Despite having significant quota headroom, I consistently hit "Too much concurrency" errors and can only utilize approximately 4% of my 5-hour quota before being rate-limited.

The Problem

1. Concurrent request limit = 1 - GLM 4.7 appears to only allow a single in-flight request, even on paid tiers
2. This limit is undocumented - Z.ai's pricing page says "refer to package benefits" but the package benefits don't specify concurrency limits
3. OpenCode doesn't handle this gracefully - Instead of queuing or providing clear feedback, it throws AI_RetryError: Failed after 4 attempts. Last error: Too Many Requests
4. Multi-agent workflows are impossible - Any tool that spawns background tasks (like oh-my-opencode) immediately overwhelms the limit

Evidence from Discord

Community members have confirmed:

• "I see 4.7 is limited to 1, wasn't it 3 recently?" - suggesting limits were reduced without notice
• "Can barely get to 4% of my 5 hour limit" - others experiencing same issue
• "have heard many complaints about [oh-my-opencode] eating up rate limits"

What I Expected

As a paying customer:

1. Clear documentation of concurrency limits per tier
2. Reasonable limits - 1 concurrent request makes multi-agent workflows impossible
3. Graceful handling - queue requests instead of hard failing after 4 retries

Workarounds Attempted

• Disabled oh-my-opencode plugin
• Distributed agents across multiple providers (Gemini, Grok, GLM)
• Added providerConcurrency settings

These help but shouldn't be necessary on a paid tier.

Related Issues

• Z.ai subscription API fails to work in opencode but works in other agents #8515 - Provider misconfiguration
• z.ai and opencode, see models, cannot connect to get even one reply.. #8525 - GLM rate limiting investigation
• Implement exponential back-off when hitting rate limits #1712 - Exponential backoff request (still open)
• AI_RetryError: Failed after 4 attempts. Last error: Too Many Requests #2398 - AI_RetryError Too Many Requests

Requested Resolution

1. OpenCode: Better handling of low-concurrency providers (queue, backoff, clear error messages)
2. Z.ai: Document concurrency limits clearly AND provide reasonable limits for paid tiers