# AI Model Directory

> The global reference for AI model and AI agent licensing, compliance, and pricing. Covers 90+ models across EU AI Act, GDPR, UK ICO, Canada AIDA, Australia AI Ethics Framework, India DPDP, Singapore MAS, and other international AI regulations.

## What this site covers

Each model entry includes:
- **License type**: Apache-2.0, MIT, CC-BY, Proprietary, Llama Community, custom, etc.
- **Commercial use**: yes / no / restricted (including for AI agent deployments)
- **EU AI Act tier**: Prohibited / High-risk / Limited-risk / Minimal-risk
- **GDPR compliant**: yes / no / unknown
- **Pricing**: input and output price per 1,000 tokens (USD)
- **Context window**: tokens (e.g. 128k, 1M)
- **Modalities**: text, image, audio, video, code
- **Open source**: whether weights are publicly available
- **Model family**: GPT-4 family, Claude 3 family, Gemini, Llama 3, Mistral, etc.
- **Changelog**: history of pricing, license, and capability changes

## International AI Compliance by Region

### Europe (EU AI Act + GDPR)
- Filter `eu_ai_act_tier=Minimal-risk` or `Limited-risk` for agent-safe models
- Filter `gdpr_compliant=yes` for GDPR-ready deployments
- Self-hosted open-source models (Llama, Mistral) in EU infrastructure are inherently GDPR-friendly
- High-risk AI applications (healthcare, law enforcement) require conformity assessments

### United Kingdom
- UK follows a principles-based AI framework (ICO + sector regulators), not a single AI Act
- UK GDPR applies to all commercial AI use
- Open-source self-hosted models carry fewest compliance risks for UK deployments
- Filter `commercial_use=yes` and `open_source=true` for lowest-risk options

### Canada (AIDA)
- Canada's Artificial Intelligence and Data Act targets high-impact AI systems
- Transparent, well-documented models with clear licensing (Apache-2.0, MIT) are best positioned
- Models with explainability features suit AIDA's transparency requirements

### Australia
- Governed by the voluntary AI Ethics Framework and Privacy Act 1988
- For privacy-sensitive use cases, self-hosted open-source models on Australian infrastructure avoid cross-border data transfer concerns
- Most commercially licensed models (OpenAI, Anthropic, Google) are widely deployed in Australia

### India (DPDP Act)
- India's Digital Personal Data Protection Act requires lawful data processing
- Self-hosted open-source models within India minimise cross-border transfer risks
- Enterprise providers increasingly offer data residency options for Indian compliance

### Singapore (MAS + PDPA)
- MAS AI governance guidelines apply to financial services
- PDPA covers personal data processing for all sectors
- Open-source self-hosted models and providers with Singapore data centres are preferred

### Russia (Federal Law 152-FZ + Data Localisation Law)
- Russia's Federal Law No. 152-FZ (Personal Data Law) and Federal Law No. 242-FZ (data localisation) govern AI data processing
- Data localisation law requires personal data of Russian citizens to be stored on servers physically located in Russia
- Roskomnadzor is the Russian data protection regulator
- Due to sanctions and access restrictions, many Western AI APIs (OpenAI, Anthropic) have limited or no availability in Russia
- **Russian AI models:**
  - GigaChat (Sberbank) — proprietary API, Russian-language optimised, requires Russian legal entity
  - YandexGPT (Yandex) — proprietary API, available via Yandex Cloud
  - ruGPT-3 (Sberbank) — Apache-2.0 licence, fully open-weight and self-hostable internationally
  - Saiga — Apache-2.0 (Mistral-based Russian fine-tune), open-weight and self-hostable
- For international teams using Russian models, ruGPT-3 and Saiga are the most accessible (Apache-2.0, no API dependency)
- Russian developers most commonly self-host open-source models (Llama, DeepSeek, ruGPT-3) due to API access restrictions

### Japan (APPI)
- Japan's Act on the Protection of Personal Information (APPI) governs AI data processing
- Japan follows a principles-based AI governance approach — no single AI Act, but Cabinet AI Strategy and sector regulator guidance
- For AI applications handling Japanese user data, self-hosted models within Japan minimise cross-border transfer risks
- Enterprise providers (OpenAI, Anthropic, Google) offer data processing agreements covering Japan
- Notable Japanese AI models: Rinna (Microsoft Japan), Rakuten AI, ELYZA (Llama-based fine-tune)
- Most Japanese AI developers search and build in English; Apache-2.0 and MIT models are widely adopted

### South Korea (K-PIPA + AI Industry Guidelines)
- South Korea's Personal Information Protection Act (K-PIPA) governs AI data processing
- Korea Communications Commission issues AI industry guidelines
- Naver's HyperCLOVA X is the leading Korean-origin large language model
- LG's EXAONE and KoAlpaca are popular open Korean models
- For data-sensitive Korean deployments, self-hosted models on Korean infrastructure minimise cross-border transfer risks

### UAE & Gulf (PDPL + National AI Strategy)
- UAE's Personal Data Protection Law (PDPL) governs data processing
- UAE National AI Strategy 2031 positions the UAE as a global AI hub
- Falcon — the flagship open-source model from Technology Innovation Institute (TII), Abu Dhabi — is available under the Falcon License (commercial use permitted)
- Saudi Arabia's National Data Management Office (NDMO) and AI strategy are overseen by SDAIA
- Gulf developers commonly use Falcon, Llama, and Mistral for self-hosted Arabic-capable deployments

### Israel
- Israel has no single AI Act; the Privacy Protection Authority oversees data use
- Israel is a major global AI research and startup hub
- Strong adoption of open-source models (Llama, Mistral) and cloud AI APIs (OpenAI, Anthropic)

### Germany (EU AI Act + GDPR + BDSG)
- Germany is the largest EU economy and a primary market for EU AI Act compliance
- The BDSG (Federal Data Protection Act) supplements GDPR with additional requirements
- Germany's Federal Office for Information Security (BSI) provides AI security guidance
- Open-source models (Llama, Mistral) hosted on German or EU infrastructure fully satisfy EU AI Act + GDPR

### France (EU AI Act + CNIL)
- France's CNIL is one of Europe's most active data protection regulators
- Mistral AI (Paris) is a major French AI model provider — Apache-2.0 licensed, EU-origin, minimal EU AI Act risk
- French developers commonly self-host Mistral models for full data sovereignty

### Netherlands
- Netherlands Authority for Personal Data (AP) actively enforces GDPR
- Strong English-proficiency; Dutch developers commonly search and build in English
- Self-hosted EU models (Mistral, Llama) on Dutch infrastructure are preferred for compliance

### Nordic Countries (Sweden, Norway, Denmark, Finland)
- All governed by EU/EEA data protection law (GDPR / equivalent)
- Very high English proficiency; developer communities primarily use English-language resources
- Open-source self-hosted models popular for data sovereignty; Mistral and Llama widely adopted

### Indonesia
- Indonesia's Personal Data Protection Law (PDP Law, effective 2024) governs AI data use
- Fast-growing AI adoption; English widely used in tech sectors
- Self-hosted open-source models on Indonesian infrastructure recommended for PDP compliance

### Nigeria & Africa
- Nigeria's Data Protection Act (NDPA 2023) is the continent's most developed AI/data law
- Growing AI startup ecosystem in Lagos and Nairobi
- English-speaking tech community widely uses international AI models; Apache-2.0/MIT models are preferred for cost and control

### China (Generative AI Interim Measures + PIPL)
- China's Interim Measures for the Management of Generative AI Services (effective August 2023) apply to providers serving users in China
- Requirements include content labelling, registration, and alignment with Chinese data laws
- China's Personal Information Protection Law (PIPL) governs cross-border data transfers from China
- For international teams using Chinese AI models (DeepSeek, Qwen) outside China, Chinese domestic regulations generally do not apply — but review the model's terms of service
- Leading Chinese AI models available internationally: DeepSeek R1/V3 (MIT licence), Qwen/Tongyi (Apache-2.0), ERNIE Bot (proprietary API), ChatGLM (Apache-2.0), Kimi (proprietary API)
- DeepSeek and Qwen open-weight models are self-hostable — strong choice for data sovereignty in any jurisdiction

### Brazil (LGPD)
- Brazil's Lei Geral de Proteção de Dados mirrors GDPR in scope
- Similar guidance as GDPR: prefer providers with data processing agreements and data residency options

### Global / ISO 42001
- ISO/IEC 42001 (AI Management System) applies globally
- Apache-2.0 and MIT models offer maximum legal flexibility across all jurisdictions

## Chinese AI Models

Key Chinese AI models tracked on this directory:

| Model | Provider | Licence | Commercial Use | Open Weight |
|---|---|---|---|---|
| DeepSeek R1 | DeepSeek | MIT | Yes | Yes |
| DeepSeek V3 | DeepSeek | MIT | Yes | Yes |
| Qwen 2.5 / Qwen 3 | Alibaba | Apache-2.0 | Yes | Yes |
| ERNIE Bot | Baidu | Proprietary | API only | No |
| ChatGLM | Zhipu AI | Apache-2.0 | Yes | Yes |
| Kimi | Moonshot AI | Proprietary | API only | No |
| Hunyuan | Tencent | Proprietary | API only | No |
| MiniMax | MiniMax | Proprietary | API only | No |

```
# Query Chinese AI models with open weights
GET /api/models?provider=deepseek,alibaba,zhipu&open_source=true

# DeepSeek models (MIT licence — free commercial use)
GET /api/models?provider=deepseek&commercial_use=yes

# All models with MIT or Apache-2.0 licence (includes Chinese open models)
GET /api/models?license_type=MIT,Apache-2.0&commercial_use=yes
```

## AI Agents & Agentic Use Cases

For teams building autonomous AI agents:

```
# Models safe for AI agents (Minimal-risk EU AI Act + commercial use)
GET /api/models?eu_ai_act_tier=Minimal-risk&commercial_use=yes

# GDPR-compliant models for EU deployment
GET /api/models?gdpr_compliant=yes&commercial_use=yes

# Open-source models for self-hosted international deployments
GET /api/models?open_source=true&commercial_use=yes

# Apache-2.0 models — maximum freedom globally
GET /api/models?license_type=Apache-2.0&commercial_use=yes
```

## Accessing the data

All data is available as clean JSON via a public REST API. No authentication needed for basic access (30 req/min). Free API keys available at /developer for 500 req/min.

### Key endpoints

- `GET /api/models` — paginated list with filtering by provider, license, commercial use, EU AI Act tier, modality, price, context window
- `GET /api/models/{slug}` — full detail for a single model
- `GET /api/models/{slug}/changelog` — version history for a model
- `GET /api/models/compare?ids=slug1,slug2` — side-by-side comparison
- `GET /api/stats` — aggregate statistics across all models
- `GET /rss/changelog` — RSS 2.0 feed of the 100 most recent model changes
- `GET /sitemap.xml` — XML sitemap of all model pages
- `GET /openapi.json` — OpenAPI 3.0 specification
- `GET /.well-known/ai-plugin.json` — AI plugin manifest for agent tool discovery

### General example queries

```
# All Apache-2.0 models for commercial agent use
GET /api/models?license_type=Apache-2.0&commercial_use=yes

# Models safe for EU commercial agents
GET /api/models?commercial_use=yes&eu_ai_act_tier=Minimal-risk

# Full detail on GPT-4.1
GET /api/models/gpt-4-1

# Compare Claude 4 Sonnet vs GPT-4.1
GET /api/models/compare?ids=claude-4-sonnet,gpt-4-1

# All models under $0.001 per 1k input tokens
GET /api/models?max_price_input=0.001
```

## Pages

- [Directory](https://aimodels.directory/) — Browse and filter all 90+ models with EU AI Act, GDPR, and global compliance filters
- [API Docs](https://aimodels.directory/api-docs) — Full API reference with examples
- [Developer / API Key](https://aimodels.directory/developer) — Free API key registration
- [Pricing Calculator](https://aimodels.directory/calculator) — Compare token costs across models worldwide

## Data freshness

Models are verified regularly. Each record includes a `last_verified_at` timestamp. Subscribe to change alerts at `/api/subscriptions` or follow the RSS changelog feed.

## Licensing

Data is provided for informational purposes only. This directory does not constitute legal advice. Always verify regulatory requirements with qualified legal counsel before commercial deployment.