Auto-Switch Control Panel
The Auto-Switch sidebar tab opens the control panel — the place to change routing
behavior. Direct URL: https://router.cocolevio.com/autoswitch/.
What you can change
| Control | Effect |
|---|---|
| Classifier | The judge model that reads prompts and picks tiers |
| SIMPLE / MEDIUM / COMPLEX / REASONING | The model pool per tier — one model id per line, top = highest priority |
The input boxes autocomplete with a curated list (ultra-cheap, free, mid, strong, and
reasoning models), but any model id the gateway knows is valid — OpenRouter ids
(openrouter/author/model), direct provider ids (gpt-4o, gemini/gemini-2.5-flash),
or Bedrock ids (bedrock/…).
Applying changes
Click Save changes. The config is written to the shared tiers.json and the router
hot-reloads it on the next request — no restart, no deploy, applies within one second.
The panel confirms with "✓ Saved — live now".
Free models: what to expect
OpenRouter :free models cost $0 but are heavily rate-limited (429s under any real
traffic). The recommended pattern, and the shipped default for SIMPLE, is free-first with
paid fallback:
openrouter/meta-llama/llama-3.3-70b-instruct:free ← $0 when available
openrouter/meta-llama/llama-3.1-8b-instruct ← $0.02/$0.03 per M when not
openrouter/google/gemini-2.5-flash-lite ← second backup
Suggested ultra-cheap models
| Model | $/M in / out |
|---|---|
openrouter/inclusionai/ling-2.6-flash | 0.01 / 0.03 |
openrouter/meta-llama/llama-3.1-8b-instruct | 0.02 / 0.03 |
openrouter/mistralai/mistral-nemo | 0.02 / 0.03 |
openrouter/amazon/nova-micro-v1 | 0.035 / 0.14 |
openrouter/qwen/qwen-2.5-7b-instruct | 0.04 / 0.10 |