Claude (Anthropic-Compatible)

POST /v1/messages — Anthropic-native chat endpoint, works directly with the Claude SDK and Claude Code.

BeLive Studio provides an Anthropic-native Messages endpoint, usable with the official Anthropic SDK or by pointing tools like Claude Code at BeLive Studio.

Endpoint

POST https://api.tokenhub.belivestudios.com/v1/messages

Anthropic's native format uses the x-api-key header, which is the recommended form shown above. The platform compatibility layer also accepts Authorization: Bearer, but Anthropic SDK and Claude Code integrations should keep x-api-key and anthropic-version.

Important note

The example above does not explicitly send a thinking configuration, but the model may still return thinking content blocks based on its capabilities. For models that support manual extended thinking, thinking.type: "enabled" makes thinking tokens count toward the turn's max_tokens limit. Under Anthropic's compatibility rules, set budget_tokens to at least 1,024 and normally below max_tokens so the final answer has room. Manual-budget mode still works on Claude 4.6 but is deprecated; newer models that support adaptive thinking should use thinking.type: "adaptive".

Common request parameters

Parameter	Type	Required	Description
`model`	string	yes	Model ID, exactly as listed in the model list
`max_tokens`	integer	yes	Maximum output tokens for the turn; thinking content also counts toward this limit
`messages`	array	yes	Conversation messages, commonly using the `user` and `assistant` roles
`system`	string \| array	no	System prompt; Anthropic's native format does not use a `system` role inside `messages`
`temperature`	number	no	Sampling temperature, usually 0–1; support depends on the model
`top_p`	number	no	Nucleus-sampling parameter; normally avoid tuning it together with `temperature`
`top_k`	integer	no	Limits candidate tokens considered for sampling; support depends on the model
`stop_sequences`	array	no	Custom stop sequences; a match normally produces `stop_reason: "stop_sequence"`
`stream`	boolean	no	Whether to return Server-Sent Events
`thinking`	object	no	Thinking configuration, such as manual `enabled` budgets or supported `adaptive` mode
`tools`	array	no	Client tool definitions; use only when tool calling is supported
`tool_choice`	object	no	Controls whether or how the model selects a tool

Parameter compatibility

Sampling parameters, thinking modes, and tool calling vary by Claude model version. Send only parameters supported by the selected model; some models restrict or ignore sampling parameters when thinking is enabled.

Extended thinking

The following Claude Opus 4.6 manual extended-thinking request was tested successfully on the platform, but this mode is deprecated on Claude 4.6. Prefer adaptive thinking for new integrations:

{
  "model": "claude-opus-4-6",
  "max_tokens": 4096,
  "thinking": {
    "type": "enabled",
    "budget_tokens": 2048
  },
  "messages": [
    {
      "role": "user",
      "content": "Analyze this problem and provide a solution"
    }
  ]
}

For models that support adaptive thinking, set thinking.type to "adaptive" and do not send budget_tokens at the same time; sending both causes a parameter error. The available thinking configuration depends on the model version.

Response fields

{
  "id": "msg_0123456789",
  "type": "message",
  "role": "assistant",
  "content": [
    {"type": "thinking", "thinking": "..."},
    {"type": "text", "text": "..."}
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 25,
    "output_tokens": 120,
    "output_tokens_details": {
      "thinking_tokens": 80
    }
  }
}

content may contain thinking and text blocks. Read each block according to its type, even when the request does not explicitly send a thinking configuration. stop_reason describes why generation ended, such as end_turn or max_tokens. usage.input_tokens and usage.output_tokens report input and output token counts, with thinking content included in output usage. When present, usage.output_tokens_details.thinking_tokens reports the thinking-token count.

Field	Description
`id`	Unique message identifier
`type`	Object type, `message` for a Messages response
`role`	Response role, normally `assistant`
`model`	Model ID actually used
`content[]`	Response content blocks, which may include `text`, `thinking`, and `tool_use`
`content[].type`	Type of the current block; parse each block according to this field
`stop_reason`	Why generation ended, such as `end_turn`, `max_tokens`, `stop_sequence`, or `tool_use`
`stop_sequence`	Custom stop sequence that was matched; normally `null` when none matched
`usage.input_tokens`	Input-token count
`usage.output_tokens`	Output-token count, including thinking consumption when enabled
`usage.output_tokens_details.thinking_tokens`	Thinking-token count when the platform response provides this detail
`usage.cache_creation_input_tokens`	Input tokens written to prompt cache; normally 0 when caching is unused
`usage.cache_read_input_tokens`	Input tokens read from prompt cache; normally 0 on a cache miss

Python SDK

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.tokenhub.belivestudios.com",
    api_key="sk-your-belive-studios-key",
)

msg = client.messages.create(
    model="claude-fable-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
text = "".join(
    block.text for block in msg.content
    if block.type == "text"
)
print(text)

Client Integration

Tools speaking the Anthropic protocol (Claude Code, CC Switch, etc.) can point their base URL at https://api.tokenhub.belivestudios.com (no /v1 — clients append the path themselves). See the Claude Code guide.