Quota & Model Routing Strategy
OriginRouter Coding Plan provides access to top-tier models under a fixed quota. The system dynamically adjusts model routing and channel selection based on the quota usage stage of your subscription tier. This strategy aims to maximize the number of available requests within the fixed quota and ensure, as much as possible, that the final request can complete without being cut off. You can enable or disable these policies at any time from the console.
This page explains how the strategy works and the available control switches.
This is a supplementary page to Coding Plan. For tier overview, supported models, and pricing, please refer to the main page.
Strategy Overview
The system classifies your subscription into usage stages based on daily / weekly quota consumption (taking the larger one), with each stage corresponding to a set of routing rules:
| Stage | Routing Behavior |
|---|---|
| Normal | Default routing by tier, preferred channels |
| Mild | More economical routing when Economy Mode is enabled |
| Moderate | Preferred channels downgraded, lighter models and fallback routes activated |
| Heavy | Forced switch to cost-effective models |
| Extreme | Top tiers retain a buffer period, targets downgraded further |
| Exhausted | Behavior depends on Balance Fallback setting |
Stage is computed on every /coding/* request in real time. Quota resets at 00:00 (UTC+8) daily or 00:00 (UTC+8) every Monday return the stage to Normal automatically.
Differences Across Tiers
Different subscription tiers have different downgrade depths at the same stage. The general patterns:
- Go tier is the most aggressive, switching to lighter models earlier when quota tightens
- Pro tier also applies downgrades under pressure, but retains slightly more channels than Go
- Max / Ultra tiers retain more buffer before entering Heavy stage, with longer availability of top models
From the model perspective:
- Top-tier models (e.g. Claude Opus series) experience the largest downgrade gap when quota tightens — this is the primary source of perceived "model capability change"
- Models that are already lightweight or use proprietary channels (e.g. the MiniMax series) remain unchanged in most stages. Kimi models are currently unavailable on the Coding endpoint.
- When Economy Mode is enabled, some previously stable lightweight models may also be routed through more cost-effective channels
Economy Mode
Economy Mode is an optional switch that voluntarily activates more economical routing earlier to extend your subscription's effective duration.
Behavior
- Normal stage: Behaves the same as when disabled
- Mild stage: A portion of requests use economical routing; when disabled, the behavior is the same as the Normal stage
- Moderate stage and above: Economical routing is forced
Recommended For
- Tasks dominated by code completion, formatting, and lightweight generation with modest intelligence requirements
- Use cases where you want to preserve quota toward the end of the cycle
Not Recommended For
- Architecture design, complex debugging, multi-file refactoring, and other tasks requiring top-model capability
This switch is enabled by default. You can toggle it at any time in the Console Subscription Settings.
Balance Fallback
Balance Fallback determines how the system behaves when quota is fully exhausted:
| Setting | Behavior at Quota Exhaustion |
|---|---|
| Enabled | Automatically deduct from account balance to continue service; console notification is sent |
| Disabled | API requests return 402 subscription_quota_exceeded; account balance is not consumed |
Important Notes
- Once Balance Fallback is enabled and quota is exhausted, all
/coding/*requests will continue consuming account balance until it is also depleted - Recommended to enable near the end of a cycle or during critical tasks; keeping it disabled under normal circumstances avoids unintended charges
Night Peak Hours
During CST night peak hours, Gemini series models automatically place the GCP channel at the front of the routing list for improved stability. This behavior is purely time-window based and independent of quota stage.
How to Determine Your Current Stage
The system does not expose a public real-time stage query interface. You can infer it indirectly:
- Sudden drop in response quality: usually indicates Moderate stage or higher
- Check usage ratio in Usage Statistics
- Receiving
402 subscription_quota_exceeded: quota is exhausted and Balance Fallback is disabled - Response quality recovers after temporarily disabling Economy Mode: likely was in Mild stage with Economy-triggered downgrades
Related Documentation
- Coding Plan — Tier overview, supported models, and pricing
- API Errors — Details of
402 subscription_quota_exceededand other errors - Usage Statistics — View real-time usage and quota consumption