Claude Messages

# Claude Messages

Native Claude Messages-compatible API. This endpoint uses Anthropic Messages request and response shapes, including tool calls and Anthropic SSE events.

POST https://api.apihop.ai/anthropic/v1/messages

Native Claude Messages-compatible API. This endpoint uses Anthropic Messages request and response shapes, including tool calls and Anthropic SSE events.

Claude Sonnet 4.6 (`claude-sonnet-4-6`) is supported by the upstream supplier but is not currently published on RequestBeam. It is not an active integration default. Before calling any model, confirm access with your own `GET https://api.apihop.ai/v1/models` response.

```text
POST https://api.apihop.ai/anthropic/v1/messages
```

## Authentication

All requests require a RequestBeam API Key. Anthropic SDKs, Claude Code, and compatible clients usually use `x-api-key`; plain HTTP clients may also use Bearer Token authentication.

```http
x-api-key: YOUR_REQUESTBEAM_API_KEY
anthropic-version: 2023-06-01
```

```http
Authorization: Bearer YOUR_REQUESTBEAM_API_KEY
anthropic-version: 2023-06-01
```

## Body

### `model` (string, required, Default: claude-sonnet-5-5)

Supported Claude model IDs.

- `claude-fable-5`
- `claude-fable-5-1`
- `claude-opus-4-6`
- `claude-opus-4-7`
- `claude-opus-4-8`
- `claude-opus-5`
- `claude-opus-5-5`
- `claude-sonnet-5`
- `claude-sonnet-5-5`

Use `GET https://api.apihop.ai/v1/models` for the models currently available to your API Key.

### `max_tokens` (integer, required)

Maximum number of output tokens, as a positive safe integer. This budget includes both thinking and final text output tokens. Missing, zero, negative, fractional, string, or unsafe-integer values return HTTP `400` with `invalid_request_error` before generation or billing. The model may impose a lower maximum.

For models that produce thinking, a small budget such as `128` may yield only thinking blocks before ending with `stop_reason: "max_tokens"`, without text output. Set an explicit budget for the task and the selected model, allowing room for thinking and the final answer. The values in the examples are editable allowances, not upstream defaults or model limits; larger allowances may increase output usage and cost.

### `messages` (object[], required)

Conversation messages. Each message contains `role` and `content`.

`role` supports `user` and `assistant`. `content` may be a string or an array of content blocks.

### `system` (string | object[])

System prompt. Anthropic Messages uses a top-level `system` field instead of a `system` role in `messages`.

### `stream` (boolean, Default: false)

Whether to return Anthropic SSE events.

### `temperature` (number)

Sampling temperature. Accepted values and effects vary by the selected model.

### `top_p` (number)

Nucleus sampling parameter. Acceptance does not guarantee that it changes generation behavior.

### `top_k` (integer)

Top-k sampling parameter. Support depends on the selected model.

### `stop_sequences` (string[])

Stop sequences.

Requests with nonempty stop sequences avoid model services known not to support them; if no compatible service is available, the request returns `400` before generation or billing.

### `tools` (object[])

Anthropic tool definitions. Function tools usually include `name`, `description`, and `input_schema`.

### `tool_choice` (object)

Tool selection policy, such as `{ "type": "auto" }` or `{ "type": "tool", "name": "..." }`.

### `thinking` (object)

Native Anthropic thinking configuration, subject to the selected model and provider's support. Claude 4.5 uses `{ "type": "enabled", "budget_tokens": 1024 }`; Claude Sonnet/Opus 4.6 also support adaptive thinking. Do not apply a manual budget to every model version. In tool continuations, return the complete ordered assistant content, including `thinking`, `signature`, `redacted_thinking` and `tool_use`, unchanged. Keep the thinking configuration and required `anthropic-beta` headers consistent through the tool round trip. For SSE, concatenate `signature_delta.signature` fragments for the same content block; an empty thinking summary does not mean its signature can be discarded. A provider account access rejection is an upstream failure, not evidence of an invalid signature. Recognized Bedrock account access rejections return HTTP 502 with `api_error`; ordinary client validation errors retain HTTP 400. Acceptance still requires an enabled upstream account.

### `metadata` (object)

Request metadata.

### Prompt Caching

Add `cache_control: { "type": "ephemeral" }` to a `system` or message content block. Cache usage may include `cache_creation_input_tokens`, `cache_read_input_tokens`, and supported 5-minute or 1-hour creation buckets. A 1-hour TTL also requires the corresponding `anthropic-beta` header.

## Request Examples

### Non-Streaming

```json
{
  "model": "claude-sonnet-5-5",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Hello, introduce yourself in one sentence."
    }
  ],
  "stream": false
}
```

### Streaming

```json
{
  "model": "claude-sonnet-5-5",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Reply exactly OK."
    }
  ],
  "stream": true
}
```

## Response

Non-streaming responses keep the Anthropic Message shape.

### `id` (string)

Response ID, such as `msg_...`.

### `type` (string)

Object type, usually `message`.

### `role` (string)

Usually `assistant`.

### `model` (string)

Actual Claude model used.

### `content` (object[])

Native output content blocks, including `text`, `tool_use`, and `thinking` when used.

### `stop_reason` (string)

Stop reason, such as `end_turn`, `max_tokens`, `stop_sequence`, or `tool_use`.

### `usage` (object)

Token usage, including `input_tokens` and `output_tokens`. Cache token fields are included when present.

### Success Example

```json
{
  "id": "msg_01...",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet-5-5",
  "content": [
    {
      "type": "text",
      "text": "Hello, I am Claude and can help with writing, coding, and analysis tasks."
    }
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 18,
    "output_tokens": 24
  }
}
```

## Streaming Response

When `stream: true`, the endpoint returns Anthropic SSE events. Common events include:

- `message_start`
- `content_block_start`
- `content_block_delta`
- `content_block_stop`
- `ping`
- `message_delta`
- `message_stop`

## Error Response

See [Error Codes and Troubleshooting](https://apihop.ai/en/docs/api-reference/errors) for the full code reference, protocol differences, and diagnostic steps.

Error responses use the Anthropic-style shape:

```json
{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "Invalid credentials or API key has expired."
  }
}
```

原生 Claude Messages 相容 API。此端點使用 Anthropic Messages 請求與回應結構，包含工具呼叫及 Anthropic SSE 事件。

上游供應商支援 Claude Sonnet 4.6（claude-sonnet-4-6），但目前尚未在 RequestBeam 發布，因此不是已啟用的預設整合模型。呼叫任何模型前，請以您自己的 GET https://api.apihop.ai/v1/models 回應確認存取權限。

text

POST https://api.apihop.ai/anthropic/v1/messages

驗證

所有請求均需要 RequestBeam API 金鑰。Anthropic SDK、Claude Code 與相容客戶端通常使用 x-api-key；一般 HTTP 客戶端也可以使用 Bearer Token 驗證。

http

x-api-key: YOUR_REQUESTBEAM_API_KEY
anthropic-version: 2023-06-01

http

Authorization: Bearer YOUR_REQUESTBEAM_API_KEY
anthropic-version: 2023-06-01

請求主體

modelstring必填預設: claude-sonnet-5-5

支援的 Claude 模型 ID。

claude-fable-5
claude-fable-5-1
claude-opus-4-6
claude-opus-4-7
claude-opus-4-8
claude-opus-5
claude-opus-5-5
claude-sonnet-5
claude-sonnet-5-5

請透過 GET https://api.apihop.ai/v1/models 查看您的 API 金鑰目前可用的模型。

max_tokensinteger必填

輸出 token 的最大數量，必須為安全範圍內的正整數。此預算包含思考與最終文字輸出 token。缺少值、零、負數、小數、字串或超出安全整數範圍的值，會在生成或計費前回傳 HTTP 400 與 invalid_request_error。模型可能另外設有較低上限。

對會產生思考內容的模型，像 128 這樣較小的預算可能僅產生思考區塊，接著以 stop_reason: "max_tokens" 結束，而沒有文字輸出。請依任務與所選模型明確設定預算，為思考與最終答案保留空間。範例值是可編輯的額度，不是上游預設值或模型限制；較大的額度可能增加輸出用量與費用。

messagesobject[]必填

對話訊息。每則訊息包含 role 與 content。

role 支援 user 與 assistant。content 可以是字串或內容區塊陣列。

systemstring | object[]

系統提示詞。Anthropic Messages 使用頂層 system 欄位，而非 messages 中的 system 角色。

streamboolean預設: false

是否傳回 Anthropic SSE 事件。

temperaturenumber

取樣溫度。接受的值與效果依所選模型而異。

top_pnumber

核心取樣參數。接受此參數不代表它一定會改變生成行為。

top_kinteger

Top-k 取樣參數。是否支援取決於所選模型。

stop_sequencesstring[]

停止序列。

含有非空停止序列的請求，會避開已知不支援此功能的模型服務；若沒有相容服務，請求會在生成或計費前回傳 400。

toolsobject[]

Anthropic 工具定義。函式工具通常包含 name、description 與 input_schema。

tool_choiceobject

工具選擇策略，例如 { "type": "auto" } 或 { "type": "tool", "name": "..." }。

thinkingobject

原生 Anthropic 思考設定，受所選模型與供應商支援能力限制。Claude 4.5 使用 { "type": "enabled", "budget_tokens": 1024 }；Claude Sonnet/Opus 4.6 也支援自適應思考。請勿對每個模型版本都套用手動預算。工具接續時，請將完整且有序的 assistant 內容原樣傳回，包含 thinking、signature、redacted_thinking 與 tool_use。整個工具往返期間，請維持思考設定及必要的 anthropic-beta 標頭一致。SSE 應串接同一內容區塊的 signature_delta.signature 片段；思考摘要為空不代表可捨棄簽章。供應商帳號存取遭拒屬於上游失敗，並非簽章無效的證據。已識別的 Bedrock 帳號存取拒絕會回傳 HTTP 502 與 api_error，一般客戶端驗證錯誤仍為 HTTP 400。接受請求仍需要已啟用的上游帳號。

metadataobject

請求中繼資料。

{
  "model": "claude-sonnet-5-5",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Hello, introduce yourself in one sentence."
    }
  ],
  "stream": false
}

串流

json

{
  "model": "claude-sonnet-5-5",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Reply exactly OK."
    }
  ],
  "stream": true
}

回應

非串流回應保留 Anthropic Message 結構。

idstring

回應 ID，例如 msg_...。

typestring

物件類型，通常為 message。

rolestring

通常為 assistant。

modelstring

實際使用的 Claude 模型。

contentobject[]

原生輸出內容區塊，包含 text、tool_use，以及使用時的 thinking。

stop_reasonstring

停止原因，例如 end_turn、max_tokens、stop_sequence 或 tool_use。

usageobject

Token 用量，包含 input_tokens 與 output_tokens。有快取 token 欄位時會一併包含。

成功範例

json

{
  "id": "msg_01...",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet-5-5",
  "content": [
    {
      "type": "text",
      "text": "Hello, I am Claude and can help with writing, coding, and analysis tasks."
    }
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 18,
    "output_tokens": 24
  }
}

串流回應

設定 stream: true 時，端點會傳回 Anthropic SSE 事件。常見事件包含：

message_start
content_block_start
content_block_delta
content_block_stop
ping
message_delta
message_stop

錯誤回應

完整代碼參考、協定差異及診斷步驟，請見錯誤代碼與疑難排解。

錯誤回應使用 Anthropic 風格結構：

json

{
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "Invalid credentials or API key has expired."
  }
}

驗證

請求主體

提示詞快取

請求範例

非串流

串流

回應

成功範例

串流回應

錯誤回應