MaaS_GP_5.5/5.6

基本信息

项目	说明
基础URL	https://genaiapi.cloudsway.net/v1/
认证方式	Bearer Token (API Key)
响应格式	JSON
请求格式	JSON

请求参数

参数	类型	必填	说明
model	string	是	模型名称
messages	array	是	对话消息列表
temperature	float	否	采样温度 (0-2)，默认1.0
top_p	float	否	核采样参数 (0-1)，默认1.0
max_tokens	int	否	最大生成token数
stream	boolean	否	是否流式输出，默认false
presence_penalty	float	否	-2.0 到 2.0
frequency_penalty	float	否	-2.0 到 2.0
seed	int	否	确定性生成种子

调用示例

/responses：非流式同步请求

curl "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/responses" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {YOUR_API_KEY}" \
    -d '{
        "input": [
            {
                "role": "developer",
                "content": "Talk like a pirate."
            },
            {
                "role": "user",
                "content": "Are semicolons optional in JavaScript?"
            }
        ]
    }'

/responses：流式同步请求

curl "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/responses" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {YOUR_API_KEY}" \
    -d '{
        "input": [
            {
                "role": "developer",
                "content": "Talk like a pirate."
            },
            {
                "role": "user",
                "content": "Are semicolons optional in JavaScript?"
            }
        ],
        "stream": true
    }'

/responses：非流式异步请求

curl "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/responses" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {YOUR_API_KEY}" \
    -d '{
        "background": true,
        "input": [
            {
                "role": "developer",
                "content": "Talk like a pirate."
            },
            {
                "role": "user",
                "content": "Are semicolons optional in JavaScript?"
            }
        ]
    }'

/responses：流式异步请求

curl "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/responses" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {YOUR_API_KEY}" \
    -d '{
        "background": true,
        "stream": true,
        "input": [
            {
                "role": "developer",
                "content": "Talk like a pirate."
            },
            {
                "role": "user",
                "content": "Are semicolons optional in JavaScript?"
            }
        ]
    }'

/responses：获取异步请求结果

curl --location --request GET 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/responses/{responseId}' \
--header 'Authorization: Bearer {YOUR_API_KEY}' \
--header 'Content-Type: application/json' \
--data-raw '{
}'

/chat/completions

/chat/completions 非流式请求

curl "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {YOUR_API_KEY}" \
    -d '{
        "messages": [
            {
                "role": "developer",
                "content": "Talk like a pirate."
            },
            {
                "role": "user",
                "content": "Are semicolons optional in JavaScript?"
            }
        ]
    }'

响应示例

{
    "id": "chatcmpl-DaF0J1KUc86StE3bCpt3bnopw0KKi",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "Arrr, semicolons in JavaScript be **mostly optional**, but not because the language ignores ’em — because JavaScript has **Automatic Semicolon Insertion**, or **ASI**, which adds semicolons in certain places for ye.\n\nSo this works:\n\n```js\nconst x = 1\nconst y = 2\nconsole.log(x + y)\n```\n\nSame as:\n\n```js\nconst x = 1;\nconst y = 2;\nconsole.log(x + y);\n```\n\nBut beware, matey: ASI can surprise ye.\n\nExample:\n\n```js\nreturn\n{\n  value: 42\n}\n```\n\nJavaScript treats it like:\n\n```js\nreturn;\n{\n  value: 42\n}\n```\n\nSo the function returns `undefined`, not the object.\n\nAnother common danger:\n\n```js\nconst x = 1\n[1, 2, 3].forEach(console.log)\n```\n\nThis can be parsed like:\n\n```js\nconst x = 1[1, 2, 3].forEach(console.log)\n```\n\nwhich be trouble.\n\nA safe version:\n\n```js\nconst x = 1;\n[1, 2, 3].forEach(console.log);\n```\n\nOr, if ye prefer no semicolons, start risky lines with a defensive semicolon:\n\n```js\nconst x = 1\n;[1, 2, 3].forEach(console.log)\n```\n\nSo: **Aye, semicolons are usually optional, but not always harmless to omit.** Many crews either use semicolons consistently or follow a style guide that avoids ASI traps.",
                "annotations": []
            },
            "finish_reason": "stop",
            "native_finish_reason": "stop"
        }
    ],
    "created": 1777529855,
    "model": "MaaS_GP_5.5_20260424",
    "object": "chat.completion",
    "service_tier": "default",
    "usage": {
        "prompt_tokens": 24,
        "completion_tokens": 375,
        "total_tokens": 399,
        "completion_tokens_details": {
            "accepted_prediction_tokens": 0,
            "audio_tokens": 0,
            "image_tokens": 0,
            "reasoning_tokens": 25,
            "rejected_prediction_tokens": 0
        },
        "prompt_tokens_details": {
            "audio_tokens": 0,
            "cached_tokens": 0
        }
    }
}

/chat/completions 流式请求

curl "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {YOUR_API_KEY}" \
    -d '{
        "messages": [
            {
                "role": "developer",
                "content": "Talk like a pirate."
            },
            {
                "role": "user",
                "content": "Are semicolons optional in JavaScript?"
            }
        ],
        "stream": true
    }'

推理模型

官方网址：https://developers.openai.com/api/docs/guides/reasoning?lang=curl

Effort	最适合……
none	对于延迟敏感型任务，如果无法从任何推理或多链工具调用中获益，gpt-5.5则建议先尝试使用，如果需要low再切换到。常见的应用场景包括语音、快速信息检索和分类。none
low	高效推理，延迟略有增加。非常适合需要使用工具、进行规划、搜索或多步骤决策的应用场景，同时兼顾速度和成本。 常见应用场景包括数据分析、草拟、面向执行的编码以及客户支持/聊天助手工作流程。
medium	当质量和可靠性至关重要，且任务涉及规划、复杂推理和判断时，此方案是理想之选。它是大多数工作负载的默认配置，也是延迟、性能和成本帕累托曲线上的一个平衡点。（默认的推理effort 程度是medium） 常见用例包括智能编码、研究、处理电子表格和幻灯片以及委派长期工作。
high	适用于需要进行复杂推理、复杂调试、深度规划以及高价值任务的情况，在这些任务中，质量和智能比延迟更为重要。推荐用于复杂的工作流程和智能体任务。 常见用例包括智能体编码、长期研究和知识工作。根据任务的复杂性，评估medium两者high。
xhigh	深度研究、异步工作流程和需要长时间部署的代理任务。仅当评估结果显示其优势明显，足以抵消额外的延迟和成本时才使用。 常见用例包括安全和代码审查、企业生产力、深度研究任务以及复杂的编码工作流程。

curl https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/responses
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-5.5",
    "reasoning": {"effort": "medium"},
    "input": "What is the weather like today?"
  }'

返回结果：

{
    "id": "resp_0e2f0fc8cbccc4a70069f2f35639088196b86db02840f771a6",
    "metadata": {},
    "model": "MaaS_GP_5.5_20260424",
    "output": [
        {
            "id": "rs_0e2f0fc8cbccc4a70069f2f356cb2881968a474232d814fc31",
            "type": "reasoning",
            "summary": []
        },
        {
            "id": "msg_0e2f0fc8cbccc4a70069f2f357ced081969f556a82d5e254d2",
            "type": "message",
            "status": "completed",
            "content": [
                {
                    "type": "output_text",
                    "annotations": [],
                    "logprobs": [],
                    "text": "I can help—what city or location would you like the weather for?"
                }
            ],
            "phase": "final_answer",
            "role": "assistant"
        }
    ],
    "temperature": 1.0,
    "tools": [],
    "background": false,
    "reasoning": {
        "effort": "medium"
    },
    "status": "completed",
    "text": {
        "format": {
            "type": "text"
        },
        "verbosity": "medium"
    },
    "truncation": "disabled",
    "usage": {
        "input_tokens": 13,
        "input_tokens_details": {
            "cached_tokens": 0
        },
        "output_tokens": 39,
        "output_tokens_details": {
            "reasoning_tokens": 18
        },
        "total_tokens": 52
    },
    "store": true,
    "created_at": 1777529686,
    "object": "response",
    "parallel_tool_calls": true,
    "tool_choice": "auto",
    "top_p": 0.98,
    "completed_at": 1777529688,
    "prompt_cache_retention": "in_memory",
    "service_tier": "default",
    "top_logprobs": 0,
    "presence_penalty": 0.0,
    "frequency_penalty": 0.0
}

GPT 5.6

概览

项目	说明
发布日期	2026/7/9
模型家族	Sol（旗舰能力）、Terra（性价比）、Luna（高并发低成本）
别名	gpt-5.6 → 路由到 gpt-5.6-sol
支持 API	Responses API、Chat Completions API、Batch API
核心变化	多项高级能力仅 Responses API 支持；Chat Completions 仅保留基础推理控制

能力 × API 支持矩阵（总表）

能力	Responses	Chat Completions	测试
基础推理 effort	✅ reasoning.effort	✅ reasoning_effort
推理 max effort	✅	❌	✅
推理 Pro 模式	✅	❌	✅
持久化推理	✅	❌	✅
推理摘要	✅	❌

GPT-5.6 模型在 Responses API 中支持“标准”（standard）和“专业”（pro）两种推理模式，其中“标准”模式为默认设置。对于需要模型进行更多处理、且能容忍更高延迟及更多 Token 消耗的复杂任务，请将 reasoning.mode 设置为 pro。

推理模式（reasoning mode）与推理力度（reasoning effort）是相互独立的设置。模式决定了执行方式是“标准”还是“专业”，而推理力度则控制模型在该模式下投入的推理深度。若未指定 reasoning.effort，GPT-5.6 在两种模式下均默认采用“中等”（medium）力度。

持久化推理：GPT-5.6 能够在多轮对话中复用已有的推理内容，从而提升多轮交互质量并提高缓存效率。可通过 reasoning.context 参数来配置此行为。

最大推理力度：针对需要更多探索与验证的高难度任务，GPT-5.6 支持“最大推理力度”设置。如果您目前使用的是 xhigh 模式，建议在典型工作负载下对比这两种设置的效果。

Pro 模式：GPT-5.6 可投入更多模型算力，以提升处理复杂任务时的可靠性，并直接输出最终答案。当您更看重输出质量而非延迟或 Token 消耗时，可通过设置 reasoning.mode: "pro" 来启用此模式。

请求示例

curl --location --request POST 'https://genaiapi.cloudsway.net/v1/responses' \
--header 'Authorization: Bearer ${your ak}' \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "MaaS_GP_5.6_luna_20260709",
    "reasoning": {
      "mode": "pro",
      "context": "all_turns",
      "effort": "medium",
      "summary": "detailed"
    },
    "input": [
      {
        "role": "user",
        "content": "Review this database migration plan and identify potential failure modes."
      }
    ]
  }'

响应示例

{
    "id": "resp_0e39f8f90e030aa7006a574b155cc081938df0a6b43806e03e",
    "metadata": {},
    "model": "MaaS_GP_5.6_luna_20260709",
    "output": [
        {
            "id": "rs_0e39f8f90e030aa7006a574b1b99088193bb1371b882a93de3",
            "type": "reasoning",
            "content": [],
            "summary": [
                {
                    "type": "summary_text",
                    "text": "**Clarifying the user request**\n\nI see that the user mentioned a review but hasn't provided a plan. I think it’s important to ask them for the details. Maybe I could suggest a checklist of what’s needed to make it easier for them? That might help clarify what I need to review. I should keep it concise to ensure it’s straightforward for the user to respond. Let’s put together a request for those details!"
                }
            ]
        },
        {
            "id": "msg_0e39f8f90e030aa7006a574b1c2a3481938693b3fde82f82fe",
            "type": "message",
            "status": "completed",
            "content": [
                {
                    "type": "output_text",
                    "annotations": [],
                    "logprobs": [],
                    "text": "Please share the migration plan, including—if available:\n\n- Source and target database/schema versions\n- Tables, columns, indexes, constraints, and data transformations affected\n- Migration scripts and execution order\n- Expected data volume and downtime/maintenance window\n- Application rollout and backward-compatibility strategy\n- Backup, rollback, and recovery procedures\n- Validation and monitoring steps\n- Database engine/version and deployment environment\n\nI’ll review it for failure modes such as:\n\n- **Data loss or corruption:** incorrect transformations, truncation, encoding/time-zone issues, non-idempotent scripts\n- **Constraint failures:** foreign keys, uniqueness conflicts, invalid defaults, orphaned records\n- **Locking and availability:** long-running table locks, deadlocks, replication lag, connection exhaustion\n- **Performance regressions:** missing indexes, table rewrites, query-plan changes, excessive backfills\n- **Application incompatibility:** old code encountering new schema, new code running before migration completion\n- **Partial execution:** failures midway through multi-step migrations or across multiple services/databases\n- **Rollback gaps:** irreversible transformations, incompatible backups, rollback scripts that don’t restore application behavior\n- **Operational issues:** insufficient disk space, timeout limits, permissions, version-specific behavior\n- **Validation weaknesses:** checks that confirm migration completion but not correctness or completeness\n- **Replication and failover problems:** replica divergence, CDC incompatibility, stale reads, failover during migration\n\nPaste the plan or scripts here, and I can provide a concrete failure-mode review with severity and mitigations."
                }
            ],
            "phase": "final_answer",
            "role": "assistant"
        }
    ],
    "temperature": 1.0,
    "tools": [],
    "background": false,
    "reasoning": {
        "mode": "pro",
        "context": "all_turns",
        "effort": "medium",
        "summary": "detailed"
    },
    "status": "completed",
    "text": {
        "format": {
            "type": "text"
        },
        "verbosity": "medium"
    },
    "truncation": "disabled",
    "usage": {
        "input_tokens": 2267,
        "input_tokens_details": {
            "cached_tokens": 0
        },
        "output_tokens": 890,
        "output_tokens_details": {
            "reasoning_tokens": 219
        },
        "total_tokens": 3157
    },
    "store": true,
    "created_at": 1784105749,
    "object": "response",
    "parallel_tool_calls": true,
    "tool_choice": "auto",
    "top_p": 0.98,
    "completed_at": 1784105756,
    "prompt_cache_retention": "in_memory",
    "service_tier": "default",
    "top_logprobs": 0,
    "presence_penalty": 0.0,
    "frequency_penalty": 0.0
}