Skip to main content
Meliai gives you access to more than 60 open-weight models spanning text generation, image synthesis, audio transcription, embeddings, and reranking — all served exclusively from European infrastructure. Because every model in the catalog is open-weight, your requests are never routed to US- or APAC-based proprietary model APIs. The catalog is continuously updated as new model families are released and validated on the platform.

​
Model categories

Text generation

Chat, instruction-following, and reasoning models from Llama, Mistral, DeepSeek, Qwen, GLM, and Kimi. Suitable for chat completions, agentic pipelines, RAG, and code generation.

Image generation

Diffusion models from the FLUX family for text-to-image generation via the /v1/images/generations endpoint.

Audio / speech

Whisper for automatic speech recognition and Voxtral for advanced audio understanding, available via /v1/audio/transcriptions.

Embeddings

Dense vector embedding models for semantic search, clustering, and retrieval-augmented generation via /v1/embeddings.

Reranking

Cross-encoder reranking models that score document–query relevance for improved RAG precision via /v1/rerank.

​
Browsing and finding models

​
Hub UI

The easiest way to explore the full catalog is the Meliai Hub. You can filter by category, provider, and routing flavor, and copy model IDs directly into your code.

​
List all models via API

Fetch the live catalog programmatically with a GET /v1/models request:
The response is an OpenAI-compatible model list object. Each entry includes the model ID, supported endpoints, and metadata about available routing flavors.

​
Look up a single model

​
Model IDs

Model IDs follow the format provider/model-name, for example: Use the exact ID returned by GET /v1/models in your request’s model field.

​
Routing suffixes on model IDs

Append a colon-separated flavor suffix to any model ID to control how Meliai selects a provider: For example, qwen/qwen3-235b-a22b:price routes to the cheapest available European provider for that model. See Routing for the full breakdown.
Meliai does not serve proprietary models such as GPT-4o, Claude, or Gemini. For client compatibility, certain well-known OpenAI and Anthropic model name aliases (e.g. gpt-4o, claude-3-5-sonnet) are mapped to capable open-weight equivalents — your existing client code continues to work while your requests stay on European open-weight infrastructure.