OpenRouter model routing

OpenRouter model routing is picking a primary model slug plus optional models fallback list so cheap tiers handle routine work and capable models only run on failure or escalation.

Hub: Tokenminning with OpenRouter. Pattern: Model routing.

Fallback array pattern

Cheaper model first, capable model second:


{
  "model": "anthropic/claude-haiku-4-20250514",
  "models": ["anthropic/claude-haiku-4-20250514", "anthropic/claude-sonnet-4-5"]
}

In Python OpenAI SDK:


client.chat.completions.create(
    model="anthropic/claude-haiku-4-20250514",
    extra_body={"models": [
        "anthropic/claude-haiku-4-20250514",
        "anthropic/claude-sonnet-4-5",
    ]},
    messages=[...],
)

OpenRouter bills only the model that succeeds. Failed primary attempts on rate limits/errors fall through without charge.

Tier selection

Task	Start slug tier
Classification, grep, single-file	Haiku / flash / mini
Multi-step agent work	Sonnet / GPT-4.1 class
Deep debug, architecture	Opus / o3 / thinking — last resort

Search current pricing on openrouter.ai/models — rates change.

Provider sorting


{
  "provider": {
    "sort": "price"
  }
}

`sort`	When
`price`	Default — cost-sensitive agent loops
`throughput`	Latency-sensitive, budgeted
`latency`	Interactive UX (still cap keys)

Avoid openrouter/auto and :nitro variants until you understand your traffic mix — unpredictable tier and cost.

Wire to tools

Tool	Config surface
OpenCode	`opencode.json` model + agent overrides
Cline	Model picker + OpenRouter base URL
Aider	`.aider.conf.yml` `model: openrouter/...`

Use one OpenRouter key per tool or environment — Activity.

Cline + OpenRouter

Queries like cline openrouter usually mean wiring VS Code Cline to https://openrouter.ai/api/v1 with a dedicated key and haiku-first model list. Same fallback and cap patterns apply.

Troubleshooting

Always billed for expensive model — primary is frontier; fallbacks never run. Set primary to haiku.

Context length errors — add larger model as second models entry.

Bill spike overnight — one key, runaway loop. Lower cap; haiku-only on that key.