OpenCode Go: 11/12 models silently fail with Insufficient Balance - upstream provider accounts depleted, breaks multi-agent orchestration

[tjddnjs124]

Description

OpenCode Go has 12 models available. Since around May 14, 11 of them return "Insufficient balance" when used for subagent spawning via the Task tool. Only minimax-m2.7 works.

Steps to reproduce

1. Have an active OpenCode Go subscription (0% usage on dashboard)
2. Configure oh-my-openagent or opencode.json to use any opencode-go/ model except minimax-m2.7
3. Call task(category="quick", ...) or task(subagent_type="oracle", ...) to spawn a subagent
4. Observe the error

Expected behavior

Subagents should spawn successfully using the configured opencode-go/ model, as they did before May 14.

Actual behavior

Subagent returns: "Insufficient balance. Manage your billing here: https://opencode.ai/workspace/.../billing"

The user's Go subscription is active with 0% usage. The problem is NOT on the user's billing side.

Test results (all 12 models)

Tested ALL models under opencode-go/ provider. Only MiniMax models work:

Model	Upstream	Subagent Spawn
minimax-m2.7	MiniMax	OK
minimax-m2.5	MiniMax	OK (per #24492)
deepseek-v4-flash	DeepSeek	Insufficient balance
deepseek-v4-pro	DeepSeek	Insufficient balance
qwen3.5-plus	Qwen	Insufficient balance
qwen3.6-plus	Qwen	Insufficient balance
kimi-k2.5	Moonshot AI	Insufficient balance
kimi-k2.6	Moonshot AI	Insufficient balance
glm-5	ZhipuAI	Insufficient balance
glm-5.1	ZhipuAI	Insufficient balance
mimo-v2.5	(third-party)	Insufficient balance
mimo-v2.5-pro	(third-party)	Insufficient balance

Same pattern of failures: upstream provider accounts (DeepSeek, Moonshot, ZhipuAI, etc.) have depleted prepaid balances. This has happened repeatedly (Kimi/Moonshot in April, DeepSeek on May 14, mimo on May 17).

This is especially impactful for multi-agent orchestration workflows (Task tool -> subagents, UltraWork/Team-UltraWork modes) which are common when using orchestrator plugins like oh-my-openagent.

Suggested fix

The error message should differentiate between "your billing is exhausted" and "OpenCode Go upstream provider is temporarily unavailable". A status endpoint showing which upstream providers are degraded would also help users understand the issue without having to dig through GitHub issues.

OpenCode version

v1.15.4

Operating System

Windows

Terminal

PowerShell

Plugins

oh-my-openagent v4.1.2

[github-actions]

This issue doesn't fully meet our contributing guidelines.

What needs to be fixed:

• No issue template was used. Please use one of the three required templates: Bug Report, Feature Request, or Question
• The issue contains an AI-generated wall of text with elaborate tables, multi-section root cause analysis, and suggested fixes — long AI-generated descriptions are not acceptable per our guidelines

Please edit this issue to address the above within 2 hours, or it will be automatically closed.

---

This issue might be a duplicate of existing issues. Please check:

• Error: 402 Insufficient Balance - opencode-go #27593: Same "402 Insufficient Balance" error with opencode-go despite active subscription (open)
• OpenCode Go Kimi K2.5/K2.6 returns 429 insufficient balance despite active Go subscription #24492: OpenCode Go models returning insufficient balance while Go quota is available (open)
• [Subscriber Alert] OpenCode Go billing failure: Kimi disabled due to "insufficient balance" on proxy account #24470: OpenCode Go billing failure due to upstream proxy account balance (open)
• OpenCode Go: MiMo-V2.5-Pro returns 402 insufficient_balance despite active subscription and remaining limits #28018: OpenCode Go MiMo-V2.5-Pro same 402 insufficient_balance issue (recently closed)

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

Thanks for updating the issue! It now includes all the required information — clear reproduction steps, expected vs. actual behavior, test results, and system details. We appreciate you taking the time to fill this out properly.

[tjddnjs124]

Precision Root Cause Analysis Update

After deeper investigation, I can now explain EXACTLY why this happens and why it is subagent-specific.

Root Cause: 3-Layer Chain

1. OpenCode API returns 402 for explicit model specification
When a subagent prompt is sent with an explicit model parameter (provider=opencode-go, model=deepseek-v4-flash), the API returns 402 "Insufficient balance". If no explicit model is passed, the API uses the agent's default model and the call succeeds. This was confirmed: subagent_type="general" (no explicit model) works with deepseek, while subagent_type="oracle" (explicit model from config) fails.

2. OMO plugin disables fallback chain when config model is set
When the user configures a model in oh-my-opencode.json (explicitCategoryModel is set), the fallback chain from CATEGORY_MODEL_REQUIREMENTS is replaced with undefined. This means if the configured model fails billing, there is NO automatic retry with alternative models from the hardcoded fallback chain.

3. Config fallback models only contain the same failing model
The config's fallback_models only included the same model that already failed. So the retry loop also fails.

Why Only Certain Models

Testing all 12 opencode-go/ models:

• minimax-m2.5 / minimax-m2.7 (MiniMax upstream) -> works
• All other 10 models -> 402

This matches previous upstream depletion events:

• Apr 26: Moonshot AI (Kimi) - raw upstream error confirmed (OpenCode Go Kimi route returns Moonshot insufficient balance 429 while Go quota is available #24462)
• Apr 26: ZhipuAI (GLM) - confirmed upstream depletion (OpenCode Go Kimi K2.5/K2.6 returns 429 insufficient balance despite active Go subscription #24492)
• May 14: DeepSeek (V4 Flash/Pro) - same pattern
• May 17: MiMo - same 402 (OpenCode Go: MiMo-V2.5-Pro returns 402 insufficient_balance despite active subscription and remaining limits #28018)

Workaround

Setting agent/category models to opencode-go/minimax-m2.5 with opencode-go/deepseek-v4-flash as fallback makes everything work. Main session can still use deepseek-v4-flash directly.

[tjddnjs124]

hello?