The open-source registry of LLM models, inference providers, and provider-model mappings. Community-maintained, schema-validated, and machine-readable.
Current stats: 88 models · 42 providers · 136 mappings
OpenModels is a structured, version-controlled registry that serves as the single source of truth for LLM model metadata, inference provider details, and provider-model mappings (including pricing and rate limits). All data is stored as human-readable YAML files and validated against JSON Schemas on every pull request.
| Vendor | Models |
|---|---|
| OpenAI | GPT-4, GPT-5, GPT-5.4, GPT-5.4 Mini, GPT-5.5, GPT-5.5 Pro, GPT-OSS 20B, GPT-OSS 120B |
| Anthropic | Claude 3 Opus, Claude Haiku 4.5, Claude Sonnet 4.5, Claude Sonnet 4.6, Claude Opus 4.6, Claude Opus 4.7 |
| Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 3 Flash, Gemini 3.1 Pro, Gemini 3.1 Flash-Lite, Gemma 3 (1B/4B/12B/27B), Gemma 4 (E2B/E4B/26B/31B) | |
| xAI | Grok 4, Grok 4.1 Fast, Grok 4.20, Grok 4.3 |
| DeepSeek | DeepSeek V3, DeepSeek R1, DeepSeek V4, DeepSeek V4 Pro, DeepSeek V4 Flash |
| Meta | Llama 3.1 8B, Llama 3.2 (3B/11B/90B), Llama 3.3 70B, Llama 4 Scout, Llama 4 Maverick, Muse Spark |
| Mistral | Mistral Medium 3.5, Mistral Large 3, Mistral Small 3.1, Mistral Small 4, Devstral 2, Codestral |
| Microsoft | Phi-4, Phi-4 Mini |
| Alibaba | Qwen3 32B, Qwen3 235B, Qwen3 Coder, QwQ-32B, Qwen 3.6 (35B-A3B, 27B, Plus), Qwen 3.7 (Max, Plus) |
| NVIDIA | Nemotron 3 Super 120B, Nemotron Nano 9B |
| Moonshot | Kimi K2.6 |
| Cohere | Command A, Command R7B |
| Zhipu | GLM-4.7, GLM-5.1 |
| MiniMax | M2.7 |
| Yandex | YandexGPT 5 Lite |
| Sber | GigaChat 3.1 Ultra, GigaChat 3.1 Lightning |
| ISSAI | KazLLM 1.0 70B |
| Astana Hub | AlemLLM |
| IBM | Granite 4.1 8B, Granite 4.1 30B |
| Xiaomi | MiMo-V2.5-Pro |
| MTS AI | Cotype Nano |
| Tencent | Hy3 Preview |
| Poolside | Laguna M.1 |
| Upstage | Solar Pro 3 |
| LLM360/MBZUAI | K2 Think |
| OpenAI (Audio) | Whisper |
Alibaba Model Studio · Amazon Bedrock · Anthropic · Anyscale · Azure AI · Baseten · Cerebras · Cloudflare Workers AI · Cohere · Deep Infra · DeepSeek · Fireworks · Google AI Studio · Google (Vertex AI) · Groq · Hugging Face Inference · Hyperbolic · IBM watsonx.ai · Inference.net · Meta · MiniMax · Mistral · Modal · Moonshot · Nebius · NLP Cloud · Novita · NVIDIA NIM · OpenAI · OpenRouter · Perplexity · Replicate · SambaNova · Sber · Scaleway · SiliconFlow · Together AI · Upstage · xAI · Xiaomi MiMo · Yandex Cloud · Zhipu AI
openmodels/
├── models/ # Canonical model definitions (YAML)
├── providers/ # Inference provider definitions (YAML)
├── mappings/ # Provider-model mappings with pricing (YAML)
│ ├── anthropic/
│ ├── openai/
│ ├── together-ai/
│ └── ...
├── schemas/ # JSON Schema definitions for validation
│ ├── model.schema.json
│ ├── provider.schema.json
│ └── mapping.schema.json
├── validate_registry.py # Validation script
└── requirements.txt # Python dependencies
We welcome contributions from the community. You can add new models, providers, or mappings by opening a pull request.
- Fork this repository
- Create a new branch for your changes
- Add or update YAML files (see templates below)
- Run validation locally to check your changes
- Open a pull request
Create a new file at models/{model-id}.yaml:
id: my-new-model # Unique ID (kebab-case, lowercase)
name: My New Model # Display name
description: A brief description of the model and its capabilities.
capabilities: # What the model can do
- chat
- completion
- function-calling
- code-generation
modalities: # Input/output modalities
- text
- code
context_window: 128000 # Max context window in tokens
licensing: apache-2.0 # License (e.g., proprietary, apache-2.0, mit)
created_at: "2025-01-01T00:00:00.000Z"
updated_at: "2025-01-01T00:00:00.000Z"Required fields: id, name, description, capabilities, modalities, context_window, licensing, created_at, updated_at
Create a new file at providers/{provider-id}.yaml:
id: my-provider # Unique ID (kebab-case, lowercase)
name: My Provider # Display name
description: Description of the inference provider.
api_base_url: https://api.myprovider.com/v1
auth_type: api-key # One of: bearer, api-key, oauth2
regions: # Deployment regions
- us-east-1
- eu-west-1
compatibility: openai # API compatibility (openai, anthropic, custom)
created_at: "2025-01-01T00:00:00.000Z"
updated_at: "2025-01-01T00:00:00.000Z"Required fields: id, name, description, api_base_url, auth_type, regions, compatibility, created_at, updated_at
Create a new file at mappings/{provider-id}/{model-id}.yaml:
model_id: deepseek-v3 # Must reference an existing model
provider_id: together-ai # Must reference an existing provider
provider_model_name: deepseek-ai/DeepSeek-V3 # Provider's internal model name
pricing:
input_per_million: 0.90 # Price per 1M input tokens (required)
output_per_million: 0.90 # Price per 1M output tokens (required)
currency: USD # ISO 4217 currency code (required)
rate_limits:
requests_per_minute: 600
tokens_per_minute: 1000000
context_window_override: null # Override model's default (or null)
available_regions: # Subset of provider's regions
- us-east-1
created_at: "2025-01-01T00:00:00.000Z"
updated_at: "2025-01-01T00:00:00.000Z"Required fields: model_id, provider_id, provider_model_name, pricing, rate_limits, available_regions, created_at, updated_at
The pricing structure supports optional fields for specialized pricing dimensions:
| Field | Description |
|---|---|
input_per_million |
Price per 1M input tokens (required) |
output_per_million |
Price per 1M output tokens (required) |
currency |
ISO 4217 currency code (required) |
cache_write_per_million |
Price per 1M tokens written to cache |
cache_read_per_million |
Price per 1M tokens read from cache |
image_per_unit |
Price per image unit (vision models) |
audio_per_second |
Price per second of audio |
reasoning_tokens_per_million |
Price per 1M reasoning tokens |
search_grounding_per_query |
Price per search grounding query |
See mappings/.example-extended-pricing.yaml for a complete example.
- Use kebab-case (lowercase letters, numbers, and hyphens only)
- Must be unique across all files of the same type
- Examples:
deepseek-v3,together-ai,gpt-5-4-mini
All pull requests are automatically validated by GitHub Actions. The validation checks:
- YAML Syntax — All YAML files must be parseable
- Schema Validation — Files must conform to their JSON Schema (
schemas/) - Duplicate IDs — Model and provider IDs must be unique
- Referential Integrity — Mappings must reference existing models and providers
# Install dependencies
pip install -r requirements.txt
# Run the validation script
python validate_registry.pyExample output on success:
✅ Validation Passed
Validated 135 files successfully.
Example output on failure:
❌ Validation Failed
Summary: 2 of 135 files failed validation.
Schema Validation Errors:
models/invalid-model.yaml
- context_window: -1000 is less than the minimum of 1
JSON Schema definitions in schemas/ enforce the structure of all YAML files:
model.schema.json— Model definition schemaprovider.schema.json— Provider definition schemamapping.schema.json— Mapping definition schema
The validation workflow triggers on pull requests that modify:
models/**,providers/**,mappings/**schemas/**validate_registry.py,requirements.txt
The workflow will:
- Install Python dependencies
- Run the validation script
- Post validation results as a PR comment
- Block the PR if validation fails
On merge to main, the ingestion workflow automatically loads updated data into the platform database.
The OpenModels platform provides a REST API and web interface for querying registry data, comparing providers, and viewing telemetry (health, latency, uptime).
- API Documentation: Available at
/api/docs(Swagger UI) and/api/docs/openapi.json(OpenAPI spec) - Web Interface: Browse models, compare providers, and view real-time telemetry
- Registry Repository: github.com/openmodelsrun/openmodels
- Documentation Site: github.com/openmodelsrun/docs
This registry is open source. See the repository for license details.