smart-model-router

# Model Router

Automatically select the right LLM for any task. Stop overpaying for simple tasks. Stop under-powering complex ones.

## Quick Decision: Can You Route Without a Classifier?

Most tasks fit obvious categories. Check the **Fast Route Table** first. Only use the classifier for ambiguous cases.

### Fast Route Table

| Signal in task | Route to | Why |
|---|---|---|
| "what time" / "what date" / simple lookup | **flash** | Zero reasoning needed |
| Format conversion, CSV→JSON, extract fields | **flash** | Mechanical transformation |
| Summarize text, list bullet points | **fast** | Pattern matching, not reasoning |
| Translate text | **fast** | Well-trained capability across all models |
| Write code, implement feature, refactor | **mid** | Needs structured thinking |
| Review code, find bugs, security audit | **mid** | Analysis without deep creativity |
| Draft email, write content | **mid** | Needs tone + context awareness |
| Research + synthesize from multiple sources | **mid** | Needs breadth, not max depth |
| Debug complex system, multi-file investigation | **strong** | Needs deep reasoning chains |
| Reflect on failures, self-improvement | **strong** | Requires genuine metacognition |
| Creative writing with nuance | **strong** | Judgment + style + originality |
| Math proofs, formal logic, complex reasoning | **reasoning** | Chain-of-thought specialist |
| Architectural decisions, tradeoff analysis | **strong** | Needs weighing multiple factors |

### Tier → Model Mapping

Configure these based on your available providers:

| Tier | Default Model | Alternatives | Cost (per 100-token query) |
|---|---|---|---|
| **flash** | `gemini-flash` | `haiku` | ~$0.00007 |
| **fast** | `haiku` | `gemini-flash`, `gpt-4o-mini` | ~$0.0003 |
| **mid** | `sonnet` | `gpt-5.2`, `gemini` | ~$0.001 |
| **strong** | `opus` | `gpt-5.2-pro`, `gemini` | ~$0.003 |
| **reasoning** | `openai/o3` | `opus` (with thinking) | ~$0.002 |

## The "Good Enough" Principle

**Not ev...