AI Models on Flatkey: 160+ Official Models Explained

Flatkey gives you access to 160+ official AI models from the world’s leading providers. Every model runs on the official vendor infrastructure — Flatkey routes your request, it does not host or fine-tune models itself. You use the exact same model IDs you would use on the provider’s own API.

Supported providers

Provider	Models available	Representative models
OpenAI	12+	gpt-4o, gpt-4o-mini, gpt-5, gpt-image-2
Anthropic	8+	claude-opus-4, claude-sonnet-4, claude-haiku-4
Google	21+	gemini-2.5-pro, gemini-2.5-flash, gemini-flash-latest
DeepSeek	6+	deepseek-v3, deepseek-v3.1, deepseek-v4-pro, deepseek-v4-flash
Qwen	10+	qwen3.5-plus, qwen3.7-max, qwen3.5-flash
Z.ai / GLM	3+	glm-4.7, glm-5-turbo, glm-5.2

Endpoint families

Flatkey supports several endpoint families, each with its own request format:

OpenAI (chat completions)

POST /v1/chat/completions — used by GPT models and most non-Google providers. OpenAI-compatible JSON format.

OpenAI (responses)

POST /v1/responses — the newer OpenAI Responses API. Supports stateful multi-turn conversations.

Gemini

Gemini models can also be called via the standard /v1/chat/completions endpoint — Flatkey handles the translation automatically.

Anthropic

Claude models are accessible via the OpenAI-compatible endpoint. You can also use the native Anthropic SDK with Flatkey.

Image Generation

POST /v1/images/generations — supports gpt-image-2 and other image models.

Embeddings

POST /v1/embeddings — create text embeddings for semantic search and RAG.

Choosing a model

Every model in the Model Directory shows:

Official price: The per-million-token price the vendor charges
After-bonus price: The effective price after applying top-up bonus credits
30-day success rate: Real production traffic data, not synthetic benchmarks
Average latency: Measured on live requests through Flatkey

For most general-purpose tasks:

Cost-sensitive workloads: gpt-4o-mini, gemini-2.5-flash-lite, deepseek-v4-flash, qwen3.5-flash
High-quality generation: gpt-4o, claude-sonnet-4, gemini-2.5-pro
Reasoning & research: deepseek-v3, qwen3.7-max, claude-opus-4
Image generation: gpt-image-2, gemini-3-pro-image
Embeddings: gemini-embedding-001

Model IDs

Use the model’s canonical ID exactly as listed in the Model Directory. For example:

python

response = client.chat.completions.create(
    model="claude-sonnet-4-5",  # Anthropic Claude Sonnet
    messages=[{"role": "user", "content": "Summarize this document."}],
)

python

response = client.chat.completions.create(
    model="deepseek-ai/deepseek-r1",  # DeepSeek R1
    messages=[{"role": "user", "content": "Explain step by step."}],
)

When switching between providers, only the model parameter changes. All other request fields (messages, temperature, max_tokens, etc.) follow the same OpenAI-compatible schema.

Model health

Flatkey tracks live health for every model based on real production traffic:

Success rate: Percentage of requests that completed successfully in the last 30 days
Average latency: Time to first token or total response time
Volume: Total requests and tokens over the period

View real-time health data at flatkey.ai/rankings or in the Model Health reference.

​Supported providers

​Endpoint families

OpenAI (chat completions)

OpenAI (responses)

Gemini

Anthropic

Image Generation

Embeddings

​Choosing a model

​Model IDs

​Model health

Supported providers

Endpoint families

Choosing a model

Model IDs

Model health