17+ models across 3 routing paths — all reachable through the same LLM Gateway, the same connector tools, and the same governance policy, whichever model answers.
Everything else speaks the same OpenAI chat-completions format, so switching between them — or adding your own — is a config change, not a re-architecture. SCX AI is the default when you don't name a provider explicitly.
Cerebras
GPT OSS 120BBest for: General assistants · Drafting & summarising
Z.ai GLM 4.7Best for: General assistants · Drafting & summarising
A frontier reasoning model published on its own — see oxalpha.io.
Native Routing
Request these by model id and the gateway routes directly to the provider — no connector setup required if you use the shared platform key, or connect your own account for dedicated capacity and billing.
Anthropic
Claude Fable 5Best for: Complex reasoning · Long-horizon agents
Claude Opus 5Best for: Complex reasoning · Long-horizon agents
Claude Opus 4.8Best for: Complex reasoning · Long-horizon agents
Claude Opus 4.7Best for: Complex reasoning · Long-horizon agents
Claude Sonnet 5Best for: General assistants · Drafting & summarising
Claude Sonnet 4.6Best for: General assistants · Drafting & summarising
Claude Haiku 4.5Best for: High-volume tasks · Classification & extraction
Request any claude-* model id — routed directly, not through the OpenAI-compatible path.
Google Gemini
Gemini 3.6 FlashBest for: General assistants · Drafting & summarising
Gemini 3.5 FlashBest for: General assistants · Drafting & summarising
Any OpenAI-compatible endpoint works via the Custom Chat API or Local LLM connectors — point ApiSpi at infrastructure you already run, no code changes on your side.
Compare models on public benchmarks
See how models score on 18 capability benchmarks — and, for each, the features it measures and the use cases those features unlock.