Model Metadata Central

A centralized, language-agnostic, open-source approach to storing and sharing model definitions like context windows, cost per token, etc.

Note: This project does not strive to be an exhaustive registry of every model, but a registry that I can use in my own projects to provide consistent, sane defaults for model and provider selection for provider and model agnostic projects where the user can configure which models and providers they want to use.

Want a model added? Request a Model.

For an exhaustive registry, check out models.dev.

Problem

Nearly every project that wants to incorporate multiple models needs to handle knowledge about the underlying model like context window length.

This leads to the proliferation of essentially the same code repeated across multiple codebases that all need to be updated when new models are released, token costs change, a model is deprecated, etc.

Examples

• LangChain
• LiteLLM
• AutoGen
• tokentrim
• AI Research Assistant Obsidian Plugin
• AgentGPT next/src/types/modelSettings.ts
• MetaGPT utils/token_counter.py

Have more examples? Create a Pull Request.

Proposal

Centralized, language-agnostic utility that defines model information and allows developers to easily import and consume these definitions in their own codebases.

JSON Schema

A JSON Schema definition can be found in model-metadata.schema.json, and example Model Metadata definitions can be found in the /models directory. Provider definitions live in /providers with their own schema in provider.schema.json.

These schema defines properties that are relevant to the model and developers who wish to leverage it in their own codebases.

Required Properties

• model_id: The identifier of the model that the provider uses
  • Example: gpt-3.5-turbo
• model_name: The human-friendly name of the model
  • Example: GPT-3.5 Turbo
• model_type: The type of model (chat, completion, or embedding)
  • Example: chat
• context_window: The maximum number of tokens in the model's context window
  • Example: 4096

Optional Properties

• model_provider: The provider of the model in lowercase
  • Example: openai
• model_description: A human-friendly description of the model
• model_version: The version of the model
  • Example 0613
• cost_per_token: The cost per token in USD
  • Example: json { "input": 0.0000015, "output": 0.000002 }
  • Note: supports either a basic number or an object with input and output numbers to define different costs between input tokens and output tokens
• knowledge_cutoff: The training data cutoff date for the model
  • Note: This is helpful when dealing with applications where you may need to know if you should supplement the model's training data with more recent information
• tokenizer: What type of tokenization the model uses
  • family: The tokenizer family, e.g. tiktoken (OpenAI), tekken (Mistral), sentencepiece (Google), other, or unknown
    • Note: This is helpful when determining if/how the consuming application can calculate token usage using a standard open source library like tiktoken or gpt-tokenizer, etc. or an alternate approach to counting tokens
  • name: The specific tokenizer name or protocol used by the model, e.g. cl100k_base (tiktoken), p50k_base (tiktoken), v3 (tekken), etc.
• tuning: The types of tuning that the model has been given in Array format; currently supports function, instruction, code, multilingual, and multimodal
  • Example: ["function", "instruction"]
  • Note: This is helpful when deciding which models are suitable for given tasks
• deprecated: Whether the model has been deprecated by the provider
  • Example: false
  • Note: Deprecated models may still work but should be avoided for new integrations
• providers: Which providers serve this model and how to reference it on each
  • Example:

    providers:
      - provider_id: openai
        model_id_on_provider: gpt-4o
      - provider_id: openrouter
        model_id_on_provider: openai/gpt-4o

  • Note: Allows a single model to be accessed via direct provider API or aggregator. Provider definitions live in /providers.
• meta_model: If the model is a meta model, like OpenRouter's Auto Router, this field indicates that the model may have special treatment for context, tokenization, and cost

Provider Schema

Provider definitions in providers/ describe API endpoints and routing. See provider.schema.json.

Required Properties

• provider_id: Unique lowercase identifier matching the providers/ filename and used in model provider_reference
• name: Human-readable name
• api_type: One of openai_compatible, anthropic, openai, other
• routing_priority: direct (serves models itself) or aggregator (routes to other providers) or both

Provider Routing

The routing_priority field differentiates direct providers (who serve models themselves) from aggregators (who route to other providers):

routing_priority	Provider Examples
direct	OpenAI, Anthropic, DeepSeek, Groq, Mistral, Cloudflare
aggregator	OpenRouter

Ollama, LM Studio, and LocalAI use direct but default to localhost with no auth — override base_url and auth_type in your app config when deploying.

Example

provider_id: openai
name: OpenAI
website_url: https://openai.com
api_type: openai_compatible
base_url: https://api.openai.com/v1
auth_type: api_key
routing_priority: direct
status: active

Included Models

66 active models across 12 primary model providers (deprecated models exist in /models but are excluded from this list). Provider definitions cover 17 direct/local/aggregator routes.

OpenAI:

• gpt-4-turbo
• gpt-4.1
• gpt-4.1-mini
• gpt-4.1-nano
• gpt-4.5
• gpt-4o
• gpt-4o-mini
• gpt-5
• gpt-5-mini
• gpt-5-nano
• gpt-5.3-codex
• gpt-5.4
• gpt-5.4-image-2
• gpt-5.4-mini
• gpt-5.4-nano
• gpt-5.4-pro
• gpt-5.5
• gpt-5.5-pro
• o3
• o4-mini

Anthropic:

• claude-haiku-4
• claude-haiku-4-5
• claude-opus-4-6
• claude-opus-4-6-fast
• claude-opus-4-7
• claude-opus-latest
• claude-sonnet-4-2
• claude-sonnet-4-6

Google:

• gemini-2.5-flash
• gemini-2.5-pro
• gemma-4-26b-a4b-it
• gemma-4-31b-it

DeepSeek:

• deepseek-coder-v4
• deepseek-v4-flash
• deepseek-v4-pro

xAI:

• grok-4.2
• grok-4.2-multi-agent

Moonshot AI:

• kimi-k2.6
• kimi-v3

Mistral AI:

• codestral-latest
• mistral-large-2411
• mistral-large-2512
• mistral-large-latest
• mistral-medium-latest
• mistral-nemo
• mistral-small-latest

MiniMax:

• minimax-m2
• minimax-m2.1
• minimax-m2.1-highspeed
• minimax-m2.5
• minimax-m2.5-highspeed
• minimax-m2.7
• minimax-m2.7-highspeed

Z.AI:

• glm-4.7
• glm-4.7-flash
• glm-5
• glm-5-turbo
• glm-5.1
• glm-5v-turbo

Qwen:

• qwen3-32b
• qwen3.6-flash
• qwen3.6-plus

Groq:

• deepseek-r1-distill-llama-70b
• llama-3.1-8b-instant
• llama-3.3-70b-versatile

NVIDIA:

• nemotron-3-super-120b-a12b

OpenRouter:

• auto

Full definitions (including deprecated models) live in /models.

Packages

Language-specific packages wrap the registry with a typed API. See each package for usage examples.

Package	Status
TypeScript	Alpha
Python	Alpha