ChatStableClaude

Claude 4.5 Sonnet

anthropic/claude-4.5-sonnet

Streaming

Pricing

Input / 1M tokens: $0
Output / 1M tokens: $0
Example workload: $0 · 1M in + 250K out

Metered from your APIRaven balance. Normalized usage ships in every response.

Call it from any OpenAI SDK

Use the official SDK with a custom base URL. The example below is complete — swap the model id to hit a different model in the catalog.

curl https://api.apiraven.com/v1/chat/completions \
  -H "Authorization: Bearer $APIRAVEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-4.5-sonnet",
    "messages": [
      { "role": "user", "content": "Explain unified inference in one sentence." }
    ],
    "temperature": 0.7,
    "max_tokens": 512,
    "stream": false
  }'

Set stream: true for server-sent events with the same response shape.

Standardized parameters

Parameters behave the same on every backend. Values that cannot be honored are either mapped to the closest equivalent or safely dropped — never silently corrupted.

Parameter	Type	Default	Range	Status
Core
`messages` Same Chat Completions shape for every model.	array	—	—	Supported
`System prompt` First message with role: system.	string	—	—	Supported
`stream` SSE chunks are normalized to chat.completion.chunk.	boolean	false	—	Supported
Sampling
`temperature`	number	0.7	0 – 2	Supported
`top_p`	number	1	0 – 1	Supported
`top_k` Normalized to the closest backend equivalent.	integer	—	0 – 100	Supported
`frequency_penalty`	number	0	-2 – 2	Supported
`presence_penalty`	number	0	-2 – 2	Supported
Limits
`max_tokens` Aliases like max_completion_tokens are collapsed into this field.	integer	model max	1 – model max	Supported
`stop` stop_sequences and stop_token_ids are translated to this field.	string \| string[]	—	—	Supported
`seed` Deterministic output where the backend supports it.	integer	random	—	Supported
Tools & format
`response_format` json_object and json_schema are supported.	object	text	—	Supported
`tools` Three upstream tool-call formats become one OpenAI shape.	array	—	—	Supported
`tool_choice`	string \| object	auto	—	Supported
Reasoning
`reasoning_effort` Applies to reasoning models; ignored by the rest.	enum	medium	low \| medium \| high	Supported

Using Claude 4.5 Sonnet on APIRaven

Which provider actually runs my request?+

APIRaven routes each request across a pool of managed inference backends. The model id, request shape and response are identical no matter which backend serves it — you never manage provider accounts, keys or regions.

Is the API OpenAI-compatible?+

Yes. Point any OpenAI SDK at https://api.apiraven.com/v1 with your APIRaven key. Chat, vision, images, embeddings and audio all follow the OpenAI request and response shapes.

What happens when a backend degrades?+

Requests are retried and failed over automatically before the first token is returned. If a model is temporarily unavailable, the API returns one stable error envelope instead of a provider-specific failure.

How is usage billed?+

Per token for text, per image for generation and per minute for transcription — deducted from your APIRaven balance. Every response includes normalized usage so you can meter cost per request.

Ship with anthropic/claude-4.5-sonnet

One base URL, one balance, every model. Point your OpenAI SDK at APIRaven and switch models with a single string.

Get your API key