Choosing a model
Pick the right catalog model for latency, cost, and capability.
On this page
The Caedral catalog is a live, OpenRouter-compatible listing of real lab model IDs (Claude, GPT, Gemini, Llama, DeepSeek, and more), plus two Caedral-hosted specialized products. Caedral does not rebrand third-party models, and there are no proprietary chat tiers — the model ID you pass is the model ID that serves your request.
Hosted products
| Product | Model ID | What it is |
|---|---|---|
| Caedral E1 Small (embeddings) | caedral-embed-e1-small-v1 | Caedral-hosted embedding model — 384 dimensions, L2-normalized, ONNX INT8 |
| Caedral Voice V1 (speech) | caedral-voice-1 | Caedral-hosted English TTS — four voices, WAV 24 kHz mono PCM16 |
| Caedral Rerank | BAAI/bge-reranker-v2-m3 | Cross-encoder reranking served on Caedral infrastructure |
The specialized vision and voice products map to their upstream models, and that mapping is public: caedral-vision routes to google/gemini-3.1-flash-image and caedral-voice routes to openai/gpt-audio. You can call the product ID (caedral-vision, caedral-voice) or the upstream ID directly.
Decision framework
| If you need… | Start with |
|---|---|
| Cheap, fast production volume | deepseek/deepseek-v4-flash |
| Agents and tool-calling | anthropic/claude-sonnet-4.5 |
| Quality-critical frontier steps | openai/gpt-5.6-sol or anthropic/claude-opus-5 |
| Embeddings / rerank (RAG) | caedral-embed / caedral-rerank |
| English speech (Caedral-owned) | caedral-voice-1 |
Routing
function selectModel(complexity: "low" | "high") { return complexity === "high" ? "anthropic/claude-sonnet-4.5" : "deepseek/deepseek-v4-flash";} const model = selectModel(task.complexity);const response = await caedral.chat.completions.create({ model, messages: [{ role: "user", content: task.prompt }],});