TrueFoundry Models

A comprehensive, community-maintained registry of AI/LLM model configurations. This repository provides standardized model metadata including pricing, features, and token limits across all major AI providers.

Scope

This registry is metadata only, for providers the TrueFoundry LLM Gateway already integrates with. Every directory under providers/ corresponds to an existing gateway integration, and the list of those integrations lives in the gateway, not here.

That means:

• Adding or updating a model under an existing provider is exactly what this repo is for. Pricing, limits, features, modalities, deprecations — open a PR.
• Adding a new provider is not. A new providers/<name>/ directory has no effect on its own: without a gateway integration behind it, nothing reads those files at runtime. New provider support is gateway work, so open an issue describing the provider and the use case instead of sending registry YAML.

See Supported Providers below for the current list.

Why Use This?

LLM model configs change often — prices drop, features expand, limits shift. This repository provides up-to-date information across providers and makes updating stale data easy.

• Unified Schema — Consistent model configuration format across 27 providers
• Up-to-Date Pricing — Current cost information for input/output tokens, batch processing, and caching
• Feature Tracking — Know exactly what each model supports (vision, tools, structured output, etc.)
• Open Source — Community-driven updates ensure accuracy and coverage

Supported Providers

This is the complete set of providers the gateway supports. If a provider isn't listed here, the registry has no place to put its models yet. The table is generated from providers/ by update-readme-counts.ts, so don't edit it by hand.

Provider	Models	Description
OpenRouter	1072	Unified API for open source models
Microsoft Foundry	514	OpenAI and Foundry catalog models on Microsoft Foundry
Google Vertex AI	466	Gemini, PaLM on GCP
Together AI	396	Open source model hosting
DeepInfra	281	Open source model hosting
AWS Bedrock	271	Claude, Llama, Titan, Mistral on AWS
Azure OpenAI	254	OpenAI models on Azure
Azure AI Foundry	239	Azure AI models
OpenAI	164	GPT-4, GPT-4o, GPT-5, o1, o3, DALL-E, Whisper, TTS
Deepgram	145	Speech-to-text and text-to-speech models
Mistral AI	119	Mistral, Mixtral, Codestral
xAI	94	Grok models
Google Gemini	92	Gemini Pro, Ultra, Flash
Perplexity	80	Search-augmented models
Databricks	70	Databricks-hosted models
Aws Bedrock Mantle	61
Cohere	44	Command, Embed models
Anthropic	31	Claude 3, Claude 3.5, Claude 4
SambaNova	31	Enterprise AI models
Groq	24	Fast inference models
Wafer	24
Truefoundry Managed	22
Fireworks Ai	15
ElevenLabs	14	Voice synthesis and text-to-speech models
AI21	12	Jamba models
Cerebras	7	Fast inference models
Typesafe	1

Installation

Direct Clone

git clone https://github.com/truefoundry/models.git

Model Configuration Schema

Each model YAML file follows this schema (validated by CUE):

# Required
model: gpt-5.4-mini-2026-03-17        # Model identifier used by the provider's API
mode: chat                             # Primary capability (chat, embedding, image, text_to_speech, etc.)

# Optional stable foundation-model identity shared across provider deployments
canonical_model: openai/gpt-5-4-mini

# Optional verified Artificial Analysis identifiers (omit when unmatched)
external_ids:
    artificial_analysis_slug: gpt-5-4-mini
    artificial_analysis_id: 00000000-0000-0000-0000-000000000000

# Pricing — array of cost entries, each with a region ("*" for global)
costs:
    - region: "*"
      input_cost_per_token: 7.5e-7
      output_cost_per_token: 0.0000045
      cache_read_input_token_cost: 7.5e-8

# Token and context window limits
limits:
    context_window: 400000
    max_output_tokens: 128000

# Feature flags
features: [function_calling, prompt_caching, structured_output, system_messages]

# Input/output modalities
modalities:
    input: [text, image]
    output: [text]

# Extended thinking / reasoning support
thinking: true

# Documentation or pricing source URLs
sources:
    - https://developers.openai.com/api/docs/pricing

canonical_model and external_ids are optional. When set, canonical_model is <creator>/<foundation-model> (lowercase). Provider hosts such as Bedrock, Vertex, Azure, OpenRouter, and Databricks are not creators. TrueFoundry costs remain the billing source; Artificial Analysis IDs are optional cross-references only.

Set these fields on the model YAML when known and review them in the PR. CI validates identity consistency when present and re-checks Artificial Analysis IDs for changed files. See scripts/README.md and the Contributing Guide.

Directory Structure

providers/
├── <provider>/
│   ├── default.yaml        # Default params for all models under this provider
│   ├── <model>.yaml
│   └── ...

Example:

providers/
├── openai/
│   ├── default.yaml
│   ├── gpt-4o.yaml
│   ├── gpt-4o-mini.yaml
│   └── ...
├── anthropic/
│   ├── default.yaml
│   ├── claude-3-5-sonnet.yaml
│   └── ...
└── ...

Contributing

We welcome contributions! Whether it's adding a model under one of the supported providers, updating pricing, or fixing outdated information — see the Contributing Guide for details on the schema, examples, and PR process. For a provider that isn't on that list, read Scope first.

License

This project is licensed under the MIT License - see the LICENSE file for details.