Skip to content

Model Catalog

SuperQode integrates with models.dev to provide a live catalog of 5000+ models across 130+ providers. The catalog is fetched on demand, cached locally at ~/.superqode/models_cache.json with a 1-hour TTL, and serves as the single source of truth for model discovery.

Use --refresh to force re-fetch.

Model names are for reference

Model identifiers in the examples below are illustrative and change as providers release new models. Use superqode models to browse the live catalog and pick the latest model available.

Live Model Discovery

Use --live to query a provider's /v1/models endpoint directly for the freshest model list. Falls back to the cached catalog if the endpoint is unreachable. Any OpenAI-compatible endpoint works without registration.

Curated and Dynamic Providers

  • ~38 hand-curated providers in the registry with tuned definitions
  • 130+ additional providers synthesized dynamically from models.dev metadata
  • OpenAI-compatible routing for providers with api_url
  • Native LiteLLM routing for providers without api_url
  • Curated providers always take precedence over synthesized ones

CLI Usage

# Browse the catalog
superqode models
superqode models --search claude
superqode models --provider anthropic
superqode models --cap tools --free
superqode models --sort price --limit 20
superqode models --refresh
superqode models --provider ollama --live

# List providers
superqode models providers
superqode models providers --json

# Model details
superqode models show anthropic/<anthropic-balanced-model>
superqode models show <openai-model>

# Hugging Face Hub
superqode models hub qwen
superqode models hub --gguf --sort downloads
superqode models hub deepseek --mlx --limit 10

# Download models
superqode models download mlx-community/Qwen3-8B-4bit
superqode models download bartowski/Qwen2.5-Coder-7B-GGUF --to ollama --quant Q4_K_M

# Check local readiness
superqode local smoke --repo .

# Convert to MLX
superqode models convert-mlx google/gemma-4-31b-it --q-bits 8

# Manage cache
superqode models cached
superqode models rm deepseek --yes

Provider Resolution

The three-tier resolution:

  1. Hand-curated provider definition (always wins)
  2. Dynamically synthesized from models.dev metadata
  3. None (provider unknown)

For synthesized OpenAI-compatible providers, base URL can be overridden via {PROVIDER_ID}_BASE_URL env var.