Claude Sonnet 4.6 API documentation

Integrate Claude Sonnet 4.6 with your VezAPI key. The operations below share the same authentication and account balance.

Text-only requests require an integer max_tokens from 1 to 8,192. The JSON request limit is 64 KiB. Media, tools, thinking and prompt caching are currently unavailable.

Base URL: https://api.vezapi.com
Send Authorization: Bearer YOUR_VEZAPI_KEY. Use a unique Idempotency-Key for each submission; reuse it only when checking the same request.

Claude Sonnet 4.6

POST/claude/v1/messages

Model IDs: claude-sonnet-4-6

Parameter	Type	Description & constraints
`model` required	`string`	Model id for this page, e.g. `claude-sonnet-4-6`. Available ids are listed by `GET [see VezAPI documentation] Values: `"claude-sonnet-4-6"`
`messages` required	`array`	Conversation messages in chronological order. `minItems: 1`
`messages[].role` required	`string`	Values: `"user", "assistant"`
`messages[].content` required	`string`	Text content. Total JSON request size must not exceed 64 KiB.
`stream` optional	`boolean`	If set to true, the response is returned as an SSE stream. Default: `true`
`max_tokens` required	`integer`	Required output token limit, 1–8192. Only text-only requests without tools, thinking or caching are enabled. Default: `4096` `minimum: 1` `maximum: 8192`

Request example

curl -X POST 'https://api.vezapi.com/claude/v1/messages' \
  -H 'Authorization: Bearer YOUR_VEZAPI_KEY' \
  -H 'Idempotency-Key: your-unique-request-id' \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "claude-sonnet-4-6",
  "max_tokens": 256,
  "messages": [
    {
      "role": "user",
      "content": "Hello!"
    }
  ]
}'

Responses and errors

HTTP 200

Request successful.

Parameter	Type	Description & constraints
`role` optional	`string`	Returned message role
`usage` optional	`object`	Usage information returned in the response
`usage.input_tokens` optional	`integer`	Input token count
`usage.output_tokens` optional	`integer`	Output token count
`usage.cache_creation_input_tokens` optional	`integer`	Cache creation input token count
`usage.cache_read_input_tokens` optional	`integer`	Cache read input token count
`usage.service_tier` optional	`string`	Service tier
`stop_reason` optional	`string`	Reason why generation stopped
`model` optional	`string`	Actual model version returned in the response
`id` optional	`string`	Unique message identifier
`credits_consumed` optional	`number`	Credits consumed by the request
`type` optional	`string`	Top-level response object type
`content` optional	`array`	Response content blocks
`content[].input` optional	`object`	Tool input arguments
`content[].caller` optional	`object`	Tool caller metadata
`content[].caller.type` optional	`string`
`content[].name` optional	`string`	Tool name
`content[].id` optional	`string`	Tool call identifier
`content[].type` optional	`string`	Content block type

HTTP 400

Bad Request - Invalid request parameters

Parameter	Type	Description & constraints
`error` optional	`object`
`error.message` optional	`string`
`error.type` optional	`string`

HTTP 401

Unauthorized - Invalid or missing API key

Parameter	Type	Description & constraints
`error` optional	`object`
`error.message` optional	`string`
`error.type` optional	`string`

Task status and callbacks

For asynchronous tasks, keep the returned taskId. Query GET /api/v1/jobs/recordInfo?taskId=... or provide callBackUrl when creating the task. Only the account that created a task can access it or derive new resources from it.

A timeout can mean the submission was accepted. Inspect the existing request in your dashboard; do not create another task automatically. Callback delivery is retried and may arrive more than once. Deduplicate with X-VezAPI-Event-Id.

Files, logs and billing

Generated media normally follows a 14-day provider lifetime; uploaded files expire after 24 hours. Save the files you need before expiration. VezAPI streams registered files without keeping a permanent media library. Request logs are retained for two calendar months.

Credits are reserved before submission. Final deductions match actual discounted consumption one for one. Unknown usage pauses new paid requests. If available funds cannot cover final usage while protecting other reservations, settlement stays pending until funded; model usage cannot make the wallet negative. Time-based inputs require server-side metadata reads; a failed read prevents submission.