Model Select

Model Selection — Task-Aware Routing via BitRouter

You have access to 33 LLM models through BitRouter's OpenRouter upstream. Rather than using the auto load-balancer for every task, actively select the right model based on the task at hand. This saves cost and improves quality.

Decision Framework

Before starting a task, assess it along two axes:

1. Complexity: How hard is the reasoning / how many steps?
2. Cost sensitivity: Is this a one-shot or a high-volume repeated task?

Then pick the lowest tier that can handle the job well.

Model Tiers

Tier 1 — Frontier (complex reasoning, architecture, research)

Use for: multi-file refactors, system design, novel algorithm design, complex debugging, long-horizon planning, tasks where correctness is critical.

Cost guidance: $2–25/M output. Use only when task demands it.

Tier 2 — Strong (general coding, analysis, most dev work)

Use for: standard feature implementation, code review, writing tests, data analysis, technical writing, moderate-complexity debugging.

Cost guidance: $0.40–15/M output. Default tier for most work.

Tier 3 — Fast (simple tasks, high throughput, cost-sensitive)

Use for: formatting, translation, summarization, triage, simple Q&A, boilerplate generation, linting, commit messages, quick lookups.

Cost guidance: $0.13–5/M output. Use for bulk work and simple tasks.

Tier 4 — Ultra-cheap (maximum throughput, minimal cost)

Use for: batch processing, classification, extraction, format conversion, simple parsing, test data generation, any task where speed >> quality.

Cost guidance: $0.05–0.75/M output. 10–100x cheaper than frontier.

Specialist Models

These excel at specific tasks regardless of their tier placement:

How to Switch Models

Use the BitRouter CLI to create task-specific routes:

# Set a named route for the current task type
bitrouter route add my-task openrouter:anthropic/claude-sonnet-4.6

# Create a route with failover (tries in order)
bitrouter route add code-review \
  openrouter:qwen/qwen3-coder-next \
  openrouter:mistralai/devstral-2512 \
  --strategy priority

# Create a load-balanced pool for a task type
bitrouter route add batch-work \
  openrouter:openai/gpt-5-nano \
  openrouter:xiaomi/mimo-v2-flash \
  openrouter:stepfun/step-3.5-flash \
  --strategy load_balance

# Check current routes
bitrouter route list

# Remove a route when done
bitrouter route rm my-task

Then use the route name as the model in your API calls:

curl http://127.0.0.1:8787/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "code-review", "messages": [...]}'

Quick Selection Cheatsheet

Cost Comparison Examples

For a typical 2K input / 1K output task:

A 70x cost difference between cheapest and most expensive. Choose wisely.