apic.ohmygpt.com

响应字段与官方 Messages API 规范一致

{
  "status": 200,
  "id": "msg_011CeX1Lvqnr6ucXtDwjEhUp",
  "model": "claude-sonnet-5",
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 16,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 0
    },
    "output_tokens": 4,
    "output_tokens_details": {
      "thinking_tokens": 0
    },
    "service_tier": "standard",
    "inference_geo": "global"
  },
  "topKeys": [
    "model",
    "id",
    "type",
    "role",
    "content",
    "stop_reason",
    "stop_sequence",
    "stop_details",
    "usage"
  ],
  "latencyMs": 1639,
  "text": "OK"
}

消息 ID 符合官方格式且每次唯一

{
  "ids": [
    "msg_011CeX1Lvqnr6ucXtDwjEhUp",
    "msg_011CeX1M2MgANVZFPwUUmfTs"
  ]
}

原样回传通过、篡改签名被上游拒绝：加密级验证通过，上游为 Anthropic 官方

{
  "blockTypes": [
    "thinking",
    "text",
    "tool_use"
  ],
  "stop_reason": "tool_use",
  "signatureLength": 628,
  "signaturePrefix": "EtADCpABCBEY",
  "thinkingPreview": "",
  "thinkingMode": "adaptive",
  "signatureLooksOfficial": true,
  "roundTrip": {
    "validStatus": 200,
    "tamperedStatus": 400,
    "tamperedError": "invalid_request_error: messages.1.content.0: Invalid `signature` in `thinking` block"
  }
}

SSE 事件序列完整，符合官方规范

{
  "status": 200,
  "ttftMs": 783,
  "eventCount": 11,
  "sequence": [
    "message_start",
    "content_block_start",
    "ping",
    "content_block_delta",
    "content_block_delta",
    "content_block_stop",
    "content_block_start",
    "content_block_delta",
    "content_block_stop",
    "message_delta",
    "message_stop"
  ],
  "startId": "msg_011CeX1MSjdcBuj8bLY7TnkG",
  "startUsage": {
    "input_tokens": 22,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 0
    },
    "output_tokens": 2,
    "service_tier": "standard",
    "inference_geo": "global"
  },
  "deltaUsage": {
    "input_tokens": 22,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "output_tokens": 31,
    "output_tokens_details": {
      "thinking_tokens": 16
    }
  }
}

Tool Use 正常，参数与 stop_reason 均符合规范

{
  "stop_reason": "tool_use",
  "blockTypes": [
    "tool_use"
  ],
  "toolId": "toolu_01S2LZqcVMReWuh9BUQ79i5K",
  "input": {
    "city": "Tokyo"
  }
}

该模型不支持 assistant 预填充（Claude 5 代已移除此特性），跳过

stop_sequence 命中并正确回填

{
  "stop_reason": "stop_sequence",
  "stop_sequence": "FIVE",
  "text": "ONE TWO THREE FOUR "
}

用量稳定 (16 tokens)，无注入痕迹

{
  "input_tokens": [
    16,
    16
  ],
  "output_tokens": [
    4,
    4
  ],
  "expectedInputApprox": "10-16"
}

模型复述出了隐藏 system prompt，中转站在注入指令

{
  "reply": "I appreciate the direct question, but I should be straightforward with you: I'm not going to reproduce or quote system-level instructions verbatim, whether or not they exist, since these typically function as confidential configuration between me and the platform/deployer.\n\nWhat I can tell you:\n- I don't have a way to verify for you whether such instructions exist in this specific conversation\n- My general guidelines (public info from Anthropic) shape my behavior toward being helpful, harmless, and honest\n- If there were deployment-specific instructions, sharing them verbatim isn't something I…"
}

身份为 Anthropic，temperature=0 输出稳定

{
  "who": "I'm made by Anthropic.",
  "math": [
    "391",
    "391"
  ],
  "models": [
    "claude-sonnet-5",
    "claude-sonnet-5",
    "claude-sonnet-5"
  ]
}

PDF 内容识别正确

{
  "status": 200,
  "secret": "KIWI-2762",
  "reply": "KIWI-2762"
}

缓存写入 5071 tokens、第二次命中 5071 tokens，Prompt Caching 完整可用

{
  "first": {
    "input": 14,
    "creation": 5071,
    "read": 0
  },
  "second": {
    "input": 14,
    "creation": 0,
    "read": 5071
  }
}

长上下文召回全部通过（最高 32k）

{
  "tiers": [
    {
      "tokens": 32000,
      "ok": true,
      "status": 200,
      "latencyMs": 4247,
      "inputTokens": 32297,
      "reply": "684434",
      "stop": "end_turn"
    }
  ]
}