How OmniRoute's Auto Model Variants (coding, fast, cheap, smart) Differ in Routing Behavior

OmniRoute's auto/* model variants apply distinct scoring weight packs—coding for quality-first code tasks, fast for lowest latency, cheap for minimum token cost, and smart for quality with 10% exploration—to dynamically select the optimal provider-model combo at request time.

OmniRoute's Auto-Combo feature lets you request models using the auto/<variant> prefix, where the suffix determines how the routing engine scores and selects a provider-model pair. Understanding these OmniRoute auto model variants helps you optimize for latency, cost, or output quality without hardcoding specific providers.

How the Auto-Combo System Works

When you send a request with auto/coding, auto/fast, auto/cheap, or auto/smart, OmniRoute's auto-routing handler (src/sse/handlers/autoRouting.ts) performs three steps:

1. Parses the model string to extract the variant name
2. Looks up the variant in AUTO_TEMPLATE_VARIANTS (defined in open-sse/services/autoCombo/builtinCatalog.ts)
3. Creates a virtual combo via createVirtualAutoCombo and runs the 9-factor scoring engine

The selected provider-model pair is returned transparently to your application.

Variant-by-Variant Comparison

auto/coding: Quality-First for Code Generation

The auto/coding variant prioritizes task-fit and stability for programming workloads. It applies the coding weight pack, which emphasizes:

• High code completion accuracy
• Model stability for debugging tasks
• Strong performance on technical prompts

Best for: Programming, debugging, code completion, technical documentation.

curl -X POST https://your-omniroute.example.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
        "model": "auto/coding",
        "messages": [{ "role": "user", "content": "Write a Python function to compute Fibonacci numbers." }]
      }'

auto/fast: Lowest Latency Selection

The auto/fast variant minimizes response time using the ship-fast weight pack. It scores providers by inverse p95 latency plus health metrics, selecting whichever endpoint can return a token fastest.

Best for: Time-critical chat, real-time assistants, streaming applications.

curl -X POST https://your-omniroute.example.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
        "model": "auto/fast",
        "messages": [{ "role": "user", "content": "What time is it in Tokyo?" }]
      }'

auto/cheap: Minimum Token Cost

The auto/cheap variant (also aliased as auto/floor) applies the cost-saver weight pack to select the cheapest per-token provider that meets basic request requirements. It exhaustively ranks providers by price before considering quality signals.

Best for: Budget-constrained workloads, bulk processing, non-critical batch jobs.

curl -X POST https://your-omniroute.example.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
        "model": "auto/cheap",
        "messages": [{ "role": "user", "content": "Summarise this 10-page PDF." }]
      }'

auto/smart: Quality with Exploration

The auto/smart variant uses the quality-first weight pack with a 10% exploration boost. This allows the router to occasionally try newer or higher-tier models that may outperform current selections, preventing the system from getting stuck on stale optimal choices.

Best for: Creative tasks, research, cases where you want best results and are willing to accept occasional variance for discovery.

curl -X POST https://your-omniroute.example.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
        "model": "auto/smart",
        "messages": [{ "role": "user", "content": "Suggest a novel plot for a sci-fi thriller." }]
      }'

Where Variants Are Defined

The AUTO_TEMPLATE_VARIANTS map in open-sse/services/autoCombo/builtinCatalog.ts (lines 20–28 in release v3.8.50) binds each auto/* prefix to its internal variant name and weight configuration. This catalog is exposed to clients via the /v1/models endpoint through src/app/api/v1/models/catalog.ts, which returns Object.keys(AUTO_TEMPLATE_VARIANTS) for discovery.

File	Purpose
open-sse/services/autoCombo/builtinCatalog.ts	Declares AUTO_TEMPLATE_VARIANTS mapping variants to weight packs
src/sse/handlers/autoRouting.ts	Parses model strings, looks up variants, builds virtual combos
src/app/api/v1/models/catalog.ts	Exposes available auto-variants via /v1/models
src/app/api/combos/auto/route.ts	Generates virtual combo payloads for REST API requests
docs/routing/AUTO-COMBO.md	Documents variant semantics and usage examples

Summary

• auto/coding — coding weight pack for highest code quality and stability
• auto/fast — ship-fast pack for minimum p95 latency
• auto/cheap — cost-saver pack for lowest token price (alias auto/floor)
• auto/smart — quality-first pack plus 10% exploration for discovery

All four variants use the same OpenAI-compatible endpoint; only the model field changes. The router resolves the virtual combo at request time according to the selected variant's scoring weights.

Frequently Asked Questions

What happens if no provider matches an auto variant's criteria?

OmniRoute falls back to the next-best scoring provider according to the variant's weight pack. If all providers fail health checks, the request queues or returns a 503 error depending on your configuration in src/sse/handlers/autoRouting.ts.

Can I create custom auto variants beyond the built-in four?

Yes. You can extend AUTO_TEMPLATE_VARIANTS in open-sse/services/autoCombo/builtinCatalog.ts with custom weight packs, though this requires modifying the OmniRoute source and redeploying. The built-in variants cover the majority of routing needs.

Does auto/smart's 10% exploration impact latency or cost significantly?

Occasionally. The exploration boost means roughly 10% of requests may route to non-optimal providers by pure quality scores, which could be slower or pricier. The trade-off enables continuous discovery of better provider-model combinations as the ecosystem evolves.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}