Home/Docs/Model Registry

Model Registry & Provider Mapping

The Selixes Model Registry maintains standard abstract models and maps them dynamically to their respective provider endpoints. When you request a model, Selixes automatically checks capabilities, token limits, and cost profiles before routing the request.

Supported Models & Priority Matrix

Selixes categorizes models by tier to provide uniform fallbacks:

Abstract Model TierPrimary Provider TargetSecondary Fallback Target
gpt-4o (Flagship)OpenAI GPT-4oAnthropic Claude 3.5 Sonnet
gpt-4-turbo (Advanced)OpenAI GPT-4 TurboAnthropic Claude 3 Opus
gpt-4o-mini (Efficient)OpenAI GPT-4o MiniAnthropic Claude 3.5 Haiku

Dynamic Payload Translators

When a request is failover-routed, the gateway automatically handles syntax and argument differences. For example:

  • Translating max_tokens (OpenAI) to max_tokens_to_sample or max_tokens (Anthropic).
  • Wrapping system prompts correctly inside the Anthropic API top-level parameters rather than messages array.
  • Converting streaming chunk objects so that client SDK parsers don't experience decoding errors.