Reliable LLM Calls with Opper

With Opper, you get access to hundreds of LLMs from all major providers, all through a single API. Out of the box, Opper gives your applications stability and flexibility that other providers don’t.

Two features make your LLM calls more dependable while keeping your code simple:

• Automatic fallbacks → when one model fails, Opper tries the next.
• Organization-scoped aliases → a friendly name that maps to an ordered list of models.

The result? Higher success rates, fewer fire-drills, and painless migrations when models or vendors change.

How Fallbacks Work

The model field can be:

• a single model (simple)
• a list of models (with built-in fallback order)
• an alias (centralized, team-friendly, and flexible)

For structured outputs, Opper also retries JSON/XML parsing several times per model before moving on. That means higher success rates—without extra client logic.

Example Flow

Three Ways to call

1. Single model (no fallback)

{
  "name": "summarize",
  "instructions": "Summarize the following text",
  "model": "openai/gpt-4o",
  "input": {"text": "..."}
}

2. List of models (built-in-fallback)

{
  "name": "summarize",
  "instructions": "Summarize the following text",
  "model": [
    {"name": "openai/gpt-4o", "options": {"temperature": 0.3}},
    {"name": "anthropic/claude-3-5-sonnet", "options": {"temperature": 0.3}}
  ],
  "input": {"text": "..."}
}

3. Alias (team-friendly and centrally managed)

Create an alias like sonnet4 that expands to:

aws/claude-sonnet-4-eu → anthropic/claude-sonnet-4 → gcp/claude-sonnet-4

curl -X POST https://api.opper.ai/v2/models/aliases \
-H "Authorization: Bearer YOUR_OPPER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"name": "sonnet4",
"fallback_models": [
"aws/claude-sonnet-4-eu",
"anthropic/claude-sonnet-4",
"gcp/claude-sonnet-4"
],
"description": "Preferred stack: sonnet eu AWS → anthropic → gcp"
}

Then call it directly:

{
  "name": "summarize",
  "instructions": "Summarize the following text",
  "model": "sonnet4",
  "input": {"text": "..."}
}

Options and headers carry through automatically—so you configure once and reuse across all models in the alias.

Why it helps

• More successful calls → fewer failed requests for your users.
• Easier operations → no more late-night pages when a provider rate-limits or goes down.
• Safer upgrades → swap models behind an alias without touching app code.
• Cleaner code → declare preferences once; let Opper handle reliability.

Custom + Vendor models

Aliases can mix your custom models with vendor models. If a custom model is unavailable, Opper just moves to the next model in sequence—no downtime.

Built-in Observability

As with any other operation using the Opper API, every attempt is fully traced:

• Which model ran
• Structured parse retries (JSON/XML)
• Errors, if any

Use these insights to fine-tune alias order, swap models, or confirm reliability improvements.

Keep Shipping, We’ll Handle the Rest

Point your app at a single alias and stop worrying about outages, migrations, or rewrites. Opper takes care of the fallbacks—so your LLM features stay reliable, day and night.