Umwelten

Appearance

Model Discovery ​

Umwelten provides comprehensive model discovery features to help you find and compare models across all supported providers (Google, Ollama, OpenRouter, LM Studio, GitHub Models, Fireworks, MiniMax).

Listing Models ​

Basic Model List ​

bash
# List all available models
pnpm run cli -- models 

# List with JSON output for programmatic use
pnpm run cli -- models  --json

# Environment variables are loaded automatically from the nearest `.env`
pnpm run cli -- models

Filter by Provider ​

bash
# Filter by specific provider
# API keys are loaded automatically from `.env` when present
pnpm run cli -- models --provider openrouter      # requires OPENROUTER_API_KEY
pnpm run cli -- models  --provider ollama
pnpm run cli -- models  --provider google         # GOOGLE_GENERATIVE_AI_API_KEY
pnpm run cli -- models  --provider lmstudio
pnpm run cli -- models  --provider llamabarn
pnpm run cli -- models  --provider github-models  # GITHUB_TOKEN
pnpm run cli -- models  --provider fireworks      # FIREWORKS_API_KEY
pnpm run cli -- models  --provider minimax        # MINIMAX_API_KEY

# Or using pnpm cli (for local development)
pnpm run cli -- models --provider github-models

Filter by Cost ​

bash
# Show only free models
pnpm run cli -- models --free

# Sort by cost (ascending by default)
pnpm run cli -- models --sort cost

Search Models ​

bash
# Search for specific models
pnpm run cli -- models --search "gpt-4"
pnpm run cli -- models --search "gemini"
pnpm run cli -- models --search "llama"

Sorting Options ​

bash
# Sort by different fields
pnpm run cli -- models --sort addedDate --desc
pnpm run cli -- models --sort contextLength
pnpm run cli -- models --sort cost
pnpm run cli -- models --sort name

Model Information ​

Detailed Model Info ​

Get comprehensive information about a specific model:

bash
# Basic model info (finds first match across all providers)
pnpm run cli -- models --view info --id <model-id>

# Examples
pnpm run cli -- models --view info --id openai/gpt-4o
pnpm run cli -- models --view info --id gemini-3-flash-preview
pnpm run cli -- models --view info --id gemma3:12b

# Get info for a specific provider's version of a model
pnpm run cli -- models --provider github-models --view info --id openai/gpt-4.1
pnpm run cli -- models --provider openrouter --view info --id openai/gpt-4o
pnpm run cli -- models --provider google --view info --id gemini-3-flash-preview

Cost Analysis ​

View cost breakdown across all models:

bash
# View all model costs
pnpm run cli -- models costs

# Sort by different cost metrics
pnpm run cli -- models costs --sort-by prompt
pnpm run cli -- models costs --sort-by completion
pnpm run cli -- models costs --sort-by total

The costs command shows:

  • Model: Model identifier
  • Provider: Which service hosts the model (ollama, openrouter, google, github-models, fireworks, minimax, lmstudio, llamabarn, llamaswap)
  • Prompt: Cost per 1M input tokens
  • Completion: Cost per 1M output tokens
  • Total: Combined cost per 1M tokens

Models are sorted by the specified metric (default: total cost).

Understanding the Output ​

Model List Table ​

Found 150+ models

┌───────────────────┬────────────┬─────────┬───────────────┬────────────────┬────────┐
│ ID                │ Provider   │ Context │ Input Cost/1M │ Output Cost/1M │ Added  │
├───────────────────┼────────────┼─────────┼───────────────┼────────────────┼────────┤
│ openai/gpt-4o     │ openrouter │ 128K    │ $2.5000       │ $10.0000       │ 5/12/24│
│ openai/gpt-4o-mini│ openrouter │ 128K    │ $0.1500       │ $0.6000        │ 7/17/24│
│ gemini-3-flash-preview  │ google     │ 1M      │ $0.0750       │ $0.3000        │ 12/11/24│
│ gemma3:12b        │ ollama     │ 8K      │ Free          │ Free           │ 7/15/25│
└───────────────────┴────────────┴─────────┴───────────────┴────────────────┴────────┘

Key Information ​

  • ID: Model identifier used in commands
  • Provider: Which service hosts the model
  • Context: Maximum context window (8K, 128K, 1M, etc.)
  • Input/Output Cost: Cost per 1 million tokens
  • Added: When the model was added to our database

Model Selection Tips ​

For Development & Testing ​

  • Free models: Ollama models (gemma3:12b, llama3.2:latest)
  • Fast & cheap: Google Gemini 2.0 Flash, OpenAI GPT-4o-mini

For Production ​

  • High quality: OpenAI GPT-4o, Google Gemini 2.5 Pro
  • Cost-effective: Google Gemini 2.0 Flash, OpenAI GPT-4o-mini

For Specialized Tasks ​

  • Vision: Google Gemini 2.0 Flash, Ollama qwen2.5vl:latest
  • Code: Ollama codestral:latest, OpenAI GPT-4o
  • Long context: Google Gemini models (up to 2M tokens)

Provider-Specific Notes ​

Ollama ​

  • Requires local Ollama server running
  • Models must be pulled locally first: ollama pull model-name
  • Free to use (only compute costs)
  • Best for privacy and local development

Google Gemini ​

  • Excellent vision capabilities
  • Very large context windows (up to 2M tokens)
  • Competitive pricing
  • Fast inference

OpenRouter ​

  • Access to many different model providers
  • Pay-per-use pricing
  • Wide variety of models and capabilities
  • Good for testing different providers

LM Studio ​

  • Local model hosting
  • No API key required
  • Models must be downloaded and loaded in LM Studio
  • Full privacy and control

GitHub Models ​

  • Free access during preview period
  • Requires GitHub Personal Access Token with models scope
  • Access to models from OpenAI, Meta, DeepSeek, and other providers
  • OpenAI-compatible API interface
  • Models are fetched from https://models.github.ai/catalog/models
  • Put GITHUB_TOKEN in .env and run:
    bash
    pnpm run cli -- models --provider github-models

Note: GitHub Models API (models.github.ai) is different from GitHub Copilot (the IDE tool).

  • GitHub Copilot has access to Anthropic/Claude models (see GitHub Copilot supported models)
  • GitHub Models API currently does not include Anthropic models - it only includes models from OpenAI, Meta, DeepSeek, AI21 Labs, Cohere, Mistral AI, xAI, and Microsoft
  • To access Anthropic models, use the OpenRouter provider instead

Fireworks ​

  • Hosted inference via an OpenAI-compatible API
  • Good for direct access to OSS and partner-hosted models
  • Requires FIREWORKS_API_KEY
  • Use models --provider fireworks to discover currently available model IDs

MiniMax ​

  • Direct access to MiniMax models such as MiniMax-M2.5 and MiniMax-M2.1
  • Requires MINIMAX_API_KEY
  • Supports direct provider usage instead of routing through OpenRouter
  • Use models --provider minimax to inspect the currently available MiniMax catalog

Next Steps ​

Once you've discovered suitable models:

Released under the MIT License.

Copyright © 2025 The Focus AI