Anthropic Messages

POST /v1/messages  compatible  语义操作 generate  契约修订 1

状态语义：stateless; portable is default; native pins one exact Anthropic-wire provider profile and disables cross-provider fallback

查看原始契约偏差（英文）

这些是源码契约中的逐条证据。日常接入先看本页示例、字段表和 Profile 覆盖矩阵；排查兼容性边界时 再展开核对原文。

• only anthropic-version 2023-06-01 is accepted
• anthropic-beta is forwarded only in native mode and only for tokens the selected connection has been configured to accept; portable mode rejects it because the request is re-authored through the canonical model and a beta token describes the request as written
• tools are classified by execution site: Anthropic-defined client-executed tools are accepted at any dated version suffix, while provider-executed tools require the selected connection to declare provider_executed_tools, because the upstream would make network calls outside SafeTransport’s host allowlist
• family matching is anchored to <family><YYYYMMDD>, so a longer name that merely begins with a known family (bash_code_execution*) is classified on its own terms rather than as the client-executed tool it resembles
• portable mode rejects members of tools[], output_config, and message content blocks that the canonical model cannot carry, because that path re-authors the body and would otherwise drop them silently; native mode forwards them verbatim
• mcp_servers, container, and fallbacks are rejected as deliberate boundaries rather than unmodelled fields: the first two delegate egress or code execution to the upstream, and the third moves model selection outside Halro’s routing and cost attribution
• in native mode tool and content-block bodies are forwarded verbatim, so cache_control and per-tool configuration reach the provider unchanged; portable mode refuses them rather than re-authoring the body without them
• Gateway Keys are accepted through x-api-key for official SDK compatibility and are never forwarded upstream
• native mode is selected with Halro-Route-Mode and requires either the direct Anthropic or Bedrock Mantle Anthropic profile

Anthropic 兼容层。注意 base URL 不带 /v1 —— Anthropic SDK 自己会补上， 这与 OpenAI SDK 相反，接错的表现是 404。认证可用 x-api-key 或 Authorization， 但两个头同时出现且值不一致会直接判失败。

请求示例

curl

curl https://halro.example.com/v1/messages \
  -H "x-api-key: $HALRO_GATEWAY_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{ "model": "claude", "max_tokens": 256,
        "messages": [{"role": "user", "content": "你好"}] }'

Python

import os
from anthropic import Anthropic

# base_url 不带 /v1，SDK 会自己补
client = Anthropic(base_url="https://halro.example.com", api_key=os.environ["HALRO_GATEWAY_KEY"],
                   timeout=60.0, max_retries=0)
message = client.messages.create(
    model="claude", max_tokens=256,
    messages=[{"role": "user", "content": "你好"}],
)
print(message.content[0].text)

Node.js

import Anthropic from "@anthropic-ai/sdk";

// baseURL 不带 /v1，SDK 会自己补
const client = new Anthropic({
  baseURL: "https://halro.example.com", apiKey: process.env.HALRO_GATEWAY_KEY,
  timeout: 60_000, maxRetries: 0,
});
const message = await client.messages.create({
  model: "claude", max_tokens: 256,
  messages: [{ role: "user", content: "你好" }],
});
console.log(message.content[0].text);

SDK 默认会自动重试

官方 SDK 默认会重试 2 次。Halro 在发出上游请求之前就把预算预留落盘，因此客户端的 静默重试会产生多次独立的预留与结算。示例把它关掉，由你自己决定重试策略 —— 见重试、超时与幂等。

请求头

• x-api-key or Authorization
• anthropic-version
• anthropic-beta
• Halro-Route-Mode
• Content-Type

请求字段

怎么读这张表

第 2 列沉默不等于支持 —— 契约只记录已声明的限制，没列出的 Profile 是「未声明限制」， 不是「已验证支持」。第 3 列是组合级限制：某些拒绝规则在字段粒度上表达不出来 （例如某个 Profile 拒绝「带工具的流式请求」，而 tools 与 stream 单看都支持）。 字段的类型、是否必填与语义说明当前不在契约里，见页脚说明。

请求字段

字段	不支持该字段的 Provider Profile	组合级限制
model	—	1 见脚注 1
max_tokens	minimax.chat.v1、kimi.chat.v1、kimi.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1	2 见脚注 2
messages	—	3 见脚注 3
messages[].role	—	4 见脚注 4
messages[].content	—	5 见脚注 5
messages[].content[].type=document	—	6 见脚注 6
messages[].content[].type=search_result	—	7 见脚注 7
system	—	—
stream	—	8 见脚注 8
stop_sequences	openai.responses.v1、bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1、minimax.anthropic.messages.v1、minimax.chat.v1、minimax.responses.v1、kimi.chat.v1、kimi.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1	9 见脚注 9
temperature	kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、kimi.code.anthropic.messages.v1	10 见脚注 10
top_p	kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、kimi.code.anthropic.messages.v1	11 见脚注 11
top_k	openai.chat-embeddings.v1、openai.responses.v1、azure-openai.chat-embeddings.v1、deepseek.chat.v1、bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、openai-compatible.chat-embeddings.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.chat.v1、bedrock.mantle.openai.chat.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1、minimax.anthropic.messages.v1、minimax.chat.v1、minimax.responses.v1、kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1、kimi.code.anthropic.messages.v1	12 见脚注 12
tools	bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1	13 见脚注 13
tools[].type=custom	—	14 见脚注 14
tools[].type=bash_*	—	15 见脚注 15
tools[].type=text_editor_*	—	16 见脚注 16
tools[].type=memory_*	—	17 见脚注 17
tools[].type=computer_*	—	18 见脚注 18
tool_choice	bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、kimi.chat.v1、kimi.code.openai.chat.v1	19 见脚注 19
thinking	openai.chat-embeddings.v1、openai.responses.v1、azure-openai.chat-embeddings.v1、deepseek.chat.v1、bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、openai-compatible.chat-embeddings.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.chat.v1、bedrock.mantle.openai.chat.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1、minimax.chat.v1、minimax.responses.v1、kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1、kimi.code.anthropic.messages.v1	20 见脚注 20
metadata	openai.chat-embeddings.v1、openai.responses.v1、azure-openai.chat-embeddings.v1、deepseek.chat.v1、bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、openai-compatible.chat-embeddings.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.chat.v1、bedrock.mantle.openai.chat.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1、minimax.chat.v1、minimax.responses.v1、kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1、kimi.code.anthropic.messages.v1	—
service_tier	openai.chat-embeddings.v1、openai.responses.v1、azure-openai.chat-embeddings.v1、deepseek.chat.v1、bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、openai-compatible.chat-embeddings.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.chat.v1、bedrock.mantle.openai.chat.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1、minimax.chat.v1、minimax.responses.v1、kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1、kimi.code.anthropic.messages.v1	—
output_config	—	21 见脚注 21
output_config.effort	openai.responses.v1、deepseek.chat.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.responses.v1、bedrock.mantle.openai.responses.v1、bedrock.mantle.anthropic.messages.v1、minimax.anthropic.messages.v1、minimax.responses.v1、kimi.chat.v1、kimi.anthropic.messages.v1、kimi.responses.v1、kimi.code.openai.chat.v1、kimi.code.anthropic.messages.v1	22 见脚注 22
output_config.format	deepseek.chat.v1、bigmodel.cn.chat-embeddings.v1、bigmodel.global.chat.v1、bigmodel.cn.coding.chat.v1、bigmodel.global.coding.chat.v1、gemini.generate-content.text.v1beta、bedrock.runtime.converse.text.v1、bedrock.mantle.anthropic.messages.v1、minimax.anthropic.messages.v1、minimax.chat.v1、minimax.responses.v1、kimi.code.openai.chat.v1、minimax.cn.subscription.openai.chat.v1、minimax.global.subscription.openai.chat.v1、kimi.code.anthropic.messages.v1	23 见脚注 23

门面层直接拒绝的字段

这些字段在到达任何 Provider 之前就会被拒。照搬 OpenAI 文档最容易撞上的就是它们：

• provider-executed tools (web_search_*, web_fetch_*, code_execution_*, advisor_*, tool_search_*) unless the selected connection declares provider_executed_tools
• mcp_servers
• container
• fallbacks
• Anthropic-defined tools in portable mode
• strict tools in portable mode
• signed thinking in portable mode
• unknown top-level fields
• unknown members of tools[] and output_config in portable mode

响应字段

id type role content model stop_reason stop_sequence usage

流事件

message_start content_block_start content_block_delta content_block_stop message_delta message_stop ping error

Provider Profile 契约覆盖与本构建可用性

端点成熟度不提升单个 Profile

顶部的 compatible 徽章说的是端点，不是下面每一行。每行的成熟度取自该 Profile 自己的覆盖记录。

本页存在成熟度分歧

bigmodel.cn.chat-embeddings.v1 的成熟度是 experimental，低于端点的 compatible。 bigmodel.global.chat.v1 的成熟度是 experimental，低于端点的 compatible。 bigmodel.cn.coding.chat.v1 的成熟度是 experimental，低于端点的 compatible。 bigmodel.global.coding.chat.v1 的成熟度是 experimental，低于端点的 compatible。 kimi.code.openai.chat.v1 的成熟度是 experimental，低于端点的 compatible。 minimax.cn.subscription.openai.chat.v1 的成熟度是 experimental，低于端点的 compatible。 minimax.global.subscription.openai.chat.v1 的成熟度是 experimental，低于端点的 compatible。 kimi.code.anthropic.messages.v1 的成熟度是 experimental，低于端点的 compatible。

有契约，不代表本构建可创建

bedrock.runtime.converse.text.v1、kimi.responses.v1 已实现且保留在兼容性契约中，但被 Halro 08af8c8b0b10 的 served Profile matrix 隐藏，Admin 写路径也会拒绝创建。 新实例不能把这些 Profile 配置成 Deployment；已有记录只保留启动与删除通道。

Provider Profile 契约覆盖与本构建可用性

Provider Profile	成熟度	本构建	不支持的请求字段
openai.chat-embeddings.v1	compatible	可创建	metadata、service_tier、thinking、top_k
openai.responses.v1	compatible	可创建	metadata、output_config.effort、service_tier、stop_sequences、thinking、top_k
anthropic.messages.2023-06-01	compatible	可创建	未声明
anthropic.subscription.messages.v1	compatible	默认关闭：需 provider_subscriptions.anthropic_claude	未声明
azure-openai.chat-embeddings.v1	compatible	可创建	metadata、service_tier、thinking、top_k
deepseek.chat.v1	compatible	可创建	metadata、output_config.effort、output_config.format、service_tier、thinking、top_k
bigmodel.cn.chat-embeddings.v1	experimental	可创建	metadata、output_config.format、service_tier、stop_sequences、thinking、tool_choice、tools、top_k
bigmodel.global.chat.v1	experimental	可创建	metadata、output_config.format、service_tier、stop_sequences、thinking、tool_choice、tools、top_k
bigmodel.cn.coding.chat.v1	experimental	可创建	metadata、output_config.format、service_tier、stop_sequences、thinking、tool_choice、tools、top_k
bigmodel.global.coding.chat.v1	experimental	可创建	metadata、output_config.format、service_tier、stop_sequences、thinking、tool_choice、tools、top_k
openai-compatible.chat-embeddings.v1	compatible	可创建	metadata、service_tier、thinking、top_k
gemini.generate-content.text.v1beta	compatible	可创建	metadata、output_config.effort、output_config.format、service_tier、thinking、tool_choice、tools、top_k
bedrock.runtime.converse.text.v1	compatible	withheld（不可创建）	metadata、output_config.effort、output_config.format、service_tier、thinking、tool_choice、tools、top_k
bedrock.mantle.chat.v1	compatible	可创建	metadata、service_tier、thinking、top_k
bedrock.mantle.openai.chat.v1	compatible	可创建	metadata、service_tier、thinking、top_k
bedrock.mantle.responses.v1	compatible	可创建	metadata、output_config.effort、service_tier、stop_sequences、thinking、tools、top_k
bedrock.mantle.openai.responses.v1	compatible	可创建	metadata、output_config.effort、service_tier、stop_sequences、thinking、tools、top_k
bedrock.mantle.anthropic.messages.v1	compatible	可创建	output_config.effort、output_config.format
minimax.anthropic.messages.v1	compatible	可创建	top_k、stop_sequences、output_config.effort、output_config.format
minimax.chat.v1	compatible	可创建	max_tokens、metadata、output_config.format、service_tier、stop_sequences、thinking、top_k
minimax.responses.v1	compatible	可创建	metadata、output_config.effort、output_config.format、service_tier、stop_sequences、thinking、top_k
kimi.chat.v1	compatible	可创建	max_tokens、metadata、output_config.effort、service_tier、stop_sequences、temperature、thinking、tool_choice、top_k、top_p
kimi.anthropic.messages.v1	compatible	可创建	metadata、output_config.effort、service_tier、temperature、thinking、top_k、top_p
kimi.responses.v1	compatible	withheld（不可创建）	max_tokens、metadata、output_config.effort、service_tier、stop_sequences、temperature、thinking、top_k、top_p
kimi.code.openai.chat.v1	experimental	可创建	max_tokens、metadata、output_config.effort、output_config.format、service_tier、stop_sequences、temperature、thinking、top_k、top_p、tool_choice
minimax.cn.subscription.openai.chat.v1	experimental	可创建	max_tokens、metadata、output_config.format、service_tier、stop_sequences、thinking、top_k
minimax.global.subscription.openai.chat.v1	experimental	可创建	max_tokens、metadata、output_config.format、service_tier、stop_sequences、thinking、top_k
kimi.code.anthropic.messages.v1	experimental	可创建	metadata、output_config.effort、output_config.format、service_tier、temperature、thinking、top_k、top_p

已声明的变换

openai.chat-embeddings.v1

• 操作结论：Halro 会把 portable Messages content 转换到 OpenAI Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through OpenAI Chat Completions

openai.responses.v1

• 操作结论：Halro 会把 portable Messages content 转换到 the OpenAI Responses endpoint；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through the OpenAI Responses endpoint
• 操作结论：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field

anthropic.messages.2023-06-01

• 操作结论：原生模式会保留通过校验的 Anthropic content block、thinking signature 与事件；客户端应按 Anthropic 原生类型读取。 原始契约：native mode preserves validated Anthropic content blocks and events

anthropic.subscription.messages.v1

• 操作结论：原生模式会保留通过校验的 Anthropic content block、thinking signature 与事件；客户端应按 Anthropic 原生类型读取。 原始契约：native mode preserves validated Anthropic content blocks and events

azure-openai.chat-embeddings.v1

• 操作结论：Halro 会把 portable Messages content 转换到 Azure Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through Azure Chat Completions

deepseek.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 DeepSeek Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through DeepSeek Chat Completions
• 操作结论：DeepSeek 仅支持 effort 的 none/low/high 与无 schema 的 json_object；其他 effort 或 json_schema 会排除该 Profile。 原始契约：output_config.effort and output_config.format are declared unsupported at field granularity because support is value-dependent: none, low and high reach DeepSeek’s thinking switch while minimal, medium and xhigh have no rung, and DeepSeek has json_object but no schema mode
• 操作结论：便携模式不要传 Anthropic 原生 thinking；DeepSeek reasoning 只能通过 output_config.effort 表达。 原始契约：thinking stays unsupported for the same reason it is on every other portable profile — it is the Anthropic-native block config, which only native mode forwards; DeepSeek’s own thinking switch is reached through output_config.effort

bigmodel.cn.chat-embeddings.v1

• 操作结论：Halro 会把 portable Messages content 转换到 BigModel Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through BigModel Chat Completions
• 操作结论：function tools 按文档最多传递 128 个；不要超过该上限。 原始契约：function tools are carried up to the documented limit of 128
• 操作结论：采样边界、reasoning effort 和必需的 max_tokens 按字段值与确切 GLM 模型在 Provider I/O 前校验。 原始契约：sampling bounds, reasoning effort and the required max_tokens bound are checked against field values and the exact GLM model before provider I/O
• 操作结论：BigModel 没有对应的 json_schema 能力，不要提交这种格式。 原始契约：json_schema has no BigModel counterpart

bigmodel.cn.coding.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 the Coding Plan’s Chat Completions route；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through the Coding Plan’s Chat Completions route
• 操作结论：function tools 按文档最多传递 128 个；不要超过该上限。 原始契约：function tools are carried up to the documented limit of 128
• 操作结论：采样边界、reasoning effort 和必需的 max_tokens 按字段值与确切 GLM 模型在 Provider I/O 前校验。 原始契约：sampling bounds, reasoning effort and the required max_tokens bound are checked against field values and the exact GLM model before provider I/O
• 操作结论：BigModel 没有对应的 json_schema 能力，不要提交这种格式。 原始契约：json_schema has no BigModel counterpart

bigmodel.global.coding.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 the international Coding Plan’s documented Chat Completions route；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through the international Coding Plan’s documented Chat Completions route
• 操作结论：function tools 按文档最多传递 128 个；不要超过该上限。 原始契约：function tools are carried up to the documented limit of 128
• 操作结论：采样边界、reasoning effort 和必需的 max_tokens 按字段值与确切 GLM 模型在 Provider I/O 前校验。 原始契约：sampling bounds, reasoning effort and the required max_tokens bound are checked against field values and the exact GLM model before provider I/O
• 操作结论：BigModel 没有对应的 json_schema 能力，不要提交这种格式。 原始契约：json_schema has no BigModel counterpart

bigmodel.global.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 Z.AI Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through Z.AI Chat Completions
• 操作结论：function tools 按文档最多传递 128 个；不要超过该上限。 原始契约：function tools are carried up to the documented limit of 128
• 操作结论：采样边界、reasoning effort 和必需的 max_tokens 按字段值与确切 GLM 模型在 Provider I/O 前校验。 原始契约：sampling bounds, reasoning effort and the required max_tokens bound are checked against field values and the exact GLM model before provider I/O
• 操作结论：BigModel 没有对应的 json_schema 能力，不要提交这种格式。 原始契约：json_schema has no BigModel counterpart

openai-compatible.chat-embeddings.v1

• 操作结论：Halro 会把 portable Messages content 转换到 an OpenAI-compatible primitive；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through an OpenAI-compatible primitive

gemini.generate-content.text.v1beta

• 操作结论：Halro 会把 portable text Messages content 转换到 Gemini generateContent；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable text Messages content is mapped through Gemini generateContent
• 操作结论：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O

bedrock.runtime.converse.text.v1

• 操作结论：Halro 会把 portable text Messages content 转换到 Bedrock Converse；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable text Messages content is mapped through Bedrock Converse
• 操作结论：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O

bedrock.mantle.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 Bedrock Mantle Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through Bedrock Mantle Chat Completions

bedrock.mantle.openai.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 Bedrock Mantle Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through Bedrock Mantle Chat Completions

bedrock.mantle.responses.v1

• 操作结论：Halro 会把 portable Messages content 转换到 stateless Bedrock Mantle Responses；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through stateless Bedrock Mantle Responses
• 操作结论：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
• 操作结论：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry, so a request carrying it is routed away before provider I/O

bedrock.mantle.openai.responses.v1

• 操作结论：Halro 会把 portable Messages content 转换到 stateless Bedrock Mantle Responses；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through stateless Bedrock Mantle Responses
• 操作结论：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
• 操作结论：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry, so a request carrying it is routed away before provider I/O

bedrock.mantle.anthropic.messages.v1

• 操作结论：原生模式会保留通过校验的 Anthropic content block、thinking signature 与事件；客户端应按 Anthropic 原生类型读取。 原始契约：native mode preserves validated Anthropic content blocks, thinking signatures, and events
• 操作结论：便携模式必须省略 output_config；原生模式才会原样转发，且仍受构建固定的 Mantle Beta 能力边界约束。 原始契约：output_config is unsupported in portable mode only: this profile shares the Anthropic wire form and could carry the member, but its Mantle Beta capability ceiling is fixed by the build and widening it is a separate contract review, so a portable request carrying effort or format is routed away before provider I/O while native mode forwards the member verbatim

minimax.anthropic.messages.v1

• 操作结论：原生模式会保留通过校验的 Anthropic content block、thinking signature 与事件；客户端应按 Anthropic 原生类型读取。 原始契约：native mode preserves validated Anthropic content blocks, thinking signatures, and events
• 操作结论：MiniMax Messages 的原生和便携模式都不接受 top_k、stop_sequences 或 cache_control；提交前删除。 原始契约：top_k, stop_sequences and cache_control are rejected on the native path as well as the portable one, because MiniMax documents them as ignored rather than refused and a silently dropped boundary is billed as if it had been honored
• 操作结论：该 Profile 没有 Prompt Cache；不要传 cache_control，也不要把费用按缓存折扣估算。 原始契约：prompt caching is unavailable, so a cache_control marker would claim a discount that does not exist
• 操作结论：MiniMax Profile 不转发 anthropic-beta；调用 MiniMax 时不要依赖任何 Anthropic Beta 功能。 原始契约：anthropic-beta is never forwarded: MiniMax does not accept beta headers
• 操作结论：MiniMax 不接受 output_config；便携请求必须省略，未指定 reasoning 时 Halro 会用 thinking=disabled。 原始契约：output_config is unsupported because MiniMax’s reasoning switch is the older thinking member; a portable request that asks for no depth still arrives with thinking disabled, which is MiniMax’s exact spelling
• 操作结论：Messages 只接受 Anthropic 原生 block type；不要提交 MiniMax 扩展的 video 或 mid_conv_system，它们会在解码阶段被拒绝。 原始契约：video and mid_conv_system content blocks are refused by the Messages decoder, which accepts only Anthropic’s own block types — MiniMax extends the wire form and Halro does not follow it there

minimax.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 MiniMax Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through MiniMax Chat Completions
• 操作结论：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
• 操作结论：该 MiniMax Messages Profile 的 max_tokens 只在 thinking 关闭时保持“回答 Token 上限”语义；不要同时请求 reasoning。 原始契约：max_tokens is why: MiniMax has one output bound and it counts reasoning, so the answer-only bound this endpoint requires is the same quantity only while nothing is thinking

minimax.responses.v1

• 操作结论：Halro 会把 portable Messages content 转换到 stateless MiniMax Responses；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through stateless MiniMax Responses
• 操作结论：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
• 操作结论：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry

kimi.anthropic.messages.v1

• 操作结论：原生模式会保留通过校验的 Anthropic content block、thinking signature 与事件；客户端应按 Anthropic 原生类型读取。 原始契约：native mode preserves validated Anthropic content blocks, thinking signatures, and events
• 操作结论：便携模式必须省略 output_config.effort；原生模式可原样转发并读取 thinking block。 原始契约：output_config.effort is unsupported in portable mode only: a portable request naming a depth comes back carrying a thinking block, which the portable decoder refuses after the upstream has been paid, while native mode forwards the member and reads the block back verbatim
• 操作结论：output_config.format 必须携带 JSON Schema；Halro 会映射为该 Profile 的 json_schema 并要求 Provider 执行。 原始契约：output_config.format is carried: this endpoint always names a schema, and Kimi takes a json_schema through the same member and enforces it
• 操作结论：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected on both paths because Kimi pins each per model and answers any other value with an error
• 操作结论：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
• 操作结论：Kimi Profile 不转发 anthropic-beta；调用 Kimi 时不要依赖任何 Anthropic Beta 功能。 原始契约：anthropic-beta is never forwarded: Kimi documents no beta headers
• 操作结论：此面已实测执行 stop_sequences，不会像忽略该字段的 MiniMax 面一样拒绝它。 原始契约：stop_sequences is carried and was measured honored, so it is not refused the way MiniMax’s is

kimi.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 Kimi Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through Kimi Chat Completions
• 操作结论：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
• 操作结论：该 Kimi Profile 的 output_config.effort 只允许 low 或 high；medium、xhigh、max 和关闭 reasoning 都不可表示。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort and is value-dependent: low and high reach Kimi’s switch, while medium, xhigh, Anthropic’s max and a request for no reasoning at all have no Kimi rung on this profile
• 操作结论：output_config.format 必须携带 JSON Schema；Halro 会映射为该 Profile 的 json_schema 并要求 Provider 执行。 原始契约：output_config.format is carried: it is always a schema on this endpoint, and the Kimi Chat face documents the json_schema response format
• 操作结论：该 Kimi Messages Profile 仅在未请求 reasoning 时把必填 max_tokens 映射到 max_completion_tokens；带 reasoning depth 的请求不可路由。 原始契约：max_tokens is carried as max_completion_tokens, which Kimi documents as the same bound under a newer name — but the two are the same quantity only while nothing is thinking, because Kimi’s single output bound counts reasoning, measured 2026-09-01. This endpoint requires max_tokens and means the answer alone, so a request that also asks for a depth is routed away rather than having this layer decide which of the two quantities the caller meant
• 操作结论：Kimi 已实测执行 stop_sequences，最多五项、每项 32 字节；超出则在 Provider I/O 前排除目标。 原始契约：stop_sequences is carried and was measured honored, within Kimi’s published bounds of five sequences of at most 32 bytes each; a request outside them is routed away before provider I/O
• 操作结论：Kimi 与 reasoning 同用时不要指定命名工具；required 仅在 kimi-k3 可与 depth 同用，K2.x 会在预留后拒绝，生产 Route 应按模型能力拆分。 原始契约：tool_choice is value-dependent: a named function together with a depth is refused by every Kimi model and routed away before provider I/O, while required together with a depth is refused by the K2.x line alone — a per-model limit this declaration cannot express, refused after the reservation rather than costing kimi-k3 the request it serves

kimi.responses.v1

• 操作结论：Halro 会把 portable Messages content 转换到 stateless Kimi Responses；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through stateless Kimi Responses
• 操作结论：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
• 操作结论：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry
• 操作结论：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected for the same reason as on the Kimi Chat profile: Kimi models neither member
• 操作结论：output_config.format 必须携带 JSON Schema；Halro 会映射为该 Profile 的 json_schema 并要求 Provider 执行。 原始契约：output_config.format is carried: it is always a schema on this endpoint, and a schema is exactly what Kimi’s Responses face accepts

kimi.code.openai.chat.v1

• 操作结论：Halro 会把 portable Messages content 转换到 OpenAI Chat Completions；调用方继续使用北向字段，并按目标 Profile 的字段矩阵核对转换损失。 原始契约：portable Messages content is mapped through OpenAI Chat Completions
• 操作结论：Kimi Code 的 OpenAI 兼容面使用专用请求渲染器；协议形状相同不表示接受全部 OpenAI 成员。 原始契约：Kimi Code’s OpenAI-compatible face is rendered by its own request renderer, not the OpenAI one: the shared wire shape does not extend to the accepted member list
• 操作结论：未请求 reasoning depth 时明确关闭 reasoning；此产品默认推理并计费，不能依赖上游默认值。 原始契约：reasoning is switched off explicitly on every request that asks for no depth, because each model this product serves reasons by default and bills for it
• 操作结论：输出总上限包含 reasoning；只有未开启 thinking 时，才能把 max_tokens 当作纯回答上限传递。 原始契约：the single output bound counts reasoning, so an answer-only max_tokens is carried only while nothing is thinking
• 操作结论：只接受此面发布的 low/high/max 深度；上游不强制这些档位，不能假定保留深度，未发布值会被拒绝。 原始契约：the published depth ladder is low/high/max and the upstream enforces none of it, so a depth this face does not publish is refused rather than served as the product’s default depth

minimax.cn.subscription.openai.chat.v1

• 操作结论：MiniMax Subscription Access 在独立产品面复用 Chat 渲染器；按量账号的 fixture 不作为订阅证据。 原始契约：MiniMax Subscription Access reuses the OpenAI Chat wire renderer on its isolated product surface; no metered-account fixture is credited to this profile
• 操作结论：依值决定的可选字段在订阅密钥 fixture 建立证据前保持失败关闭，不能按协议相同推定支持。 原始契约：value-dependent optional fields remain fail-closed until a subscription-key fixture establishes their behavior

minimax.global.subscription.openai.chat.v1

• 操作结论：MiniMax Subscription Access 在独立产品面复用 Chat 渲染器；按量账号的 fixture 不作为订阅证据。 原始契约：MiniMax Subscription Access reuses the OpenAI Chat wire renderer on its isolated product surface; no metered-account fixture is credited to this profile
• 操作结论：依值决定的可选字段在订阅密钥 fixture 建立证据前保持失败关闭，不能按协议相同推定支持。 原始契约：value-dependent optional fields remain fail-closed until a subscription-key fixture establishes their behavior

kimi.code.anthropic.messages.v1

• 操作结论：Kimi Code 在独立会员接入面复用 Messages 渲染器；Kimi 开放平台 fixture 不作为会员产品证据。 原始契约：Kimi Code reuses the Anthropic Messages wire renderer on its isolated membership surface; no Kimi Open Platform fixture is credited to this profile
• 操作结论：可移植请求未指定深度时发送 thinking disabled；已在 k3、k3-256k、kimi-for-coding 的流式和非流式请求实测。 原始契约：thinking disabled is sent on every portable request that asks for no depth, and was measured honoured on k3, k3-256k and kimi-for-coding, non-streaming and streaming alike
• 操作结论：此面接受 temperature、top_k 并返回 200 却不能证明生效，因此 Halro 拒绝这些字段。 原始契约：temperature and top_k are refused here because this face accepts them and answers 200 without establishing that either was read, where the Chat face refuses them outright
• 操作结论：stop_sequences 已实测执行，但截断结果返回 stop_reason=end_turn 和空 stop_sequence，不要依赖命中的序列名。 原始契约：stop_sequences is carried and was measured honoured, but a cut answer reports stop_reason end_turn with a null stop_sequence rather than naming the sequence

证据

已验证的证据：gateway_contract、provider_transport_fixture、sdk_blackbox

SDK 黑盒协议桩矩阵

anthropic-go、anthropic-typescript、anthropic-python

警告

该矩阵只证明这些 SDK 版本能序列化请求、并消费 Halro 声明的响应与流形状。它跑的是协议桩， 不经过真实 Runtime 路由、Gateway Key 认证、路由选择、能力过滤、脱敏与预算核算， 不能被引为端到端网关或真实 Provider 的证据。

最佳实践

portable、native 与 Token Count 的选择见案例十。

---

本页的字段表、覆盖矩阵与偏差列表由 Halro 08af8c8b0b10 的兼容性契约生成 （scripts/generate-api-pages.mjs，当前副本摘要 db249a88ce62e5b7…）。

字段的类型、是否必填与语义说明尚未进入契约，因此本页只列字段名。

Footnotes

1. model 的组合级限制：

   • kimi.anthropic.messages.v1：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected on both paths because Kimi pins each per model and answers any other value with an error
   • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
   • kimi.chat.v1：Kimi 与 reasoning 同用时不要指定命名工具；required 仅在 kimi-k3 可与 depth 同用，K2.x 会在预留后拒绝，生产 Route 应按模型能力拆分。 原始契约：tool_choice is value-dependent: a named function together with a depth is refused by every Kimi model and routed away before provider I/O, while required together with a depth is refused by the K2.x line alone — a per-model limit this declaration cannot express, refused after the reservation rather than costing kimi-k3 the request it serves
   • kimi.responses.v1：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected for the same reason as on the Kimi Chat profile: Kimi models neither member
   ↩

2. max_tokens 的组合级限制：

   • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
   • kimi.chat.v1：该 Kimi Messages Profile 仅在未请求 reasoning 时把必填 max_tokens 映射到 max_completion_tokens；带 reasoning depth 的请求不可路由。 原始契约：max_tokens is carried as max_completion_tokens, which Kimi documents as the same bound under a newer name — but the two are the same quantity only while nothing is thinking, because Kimi’s single output bound counts reasoning, measured 2026-09-01. This endpoint requires max_tokens and means the answer alone, so a request that also asks for a depth is routed away rather than having this layer decide which of the two quantities the caller meant
   ↩

3. messages 的组合级限制：

   • openai.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • minimax.anthropic.messages.v1：Messages 只接受 Anthropic 原生 block type；不要提交 MiniMax 扩展的 video 或 mid_conv_system，它们会在解码阶段被拒绝。 原始契约：video and mid_conv_system content blocks are refused by the Messages decoder, which accepts only Anthropic’s own block types — MiniMax extends the wire form and Halro does not follow it there
   • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
   • minimax.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • kimi.anthropic.messages.v1：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
   • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
   • kimi.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   ↩

4. messages[].role 的组合级限制：

   • openai.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • minimax.anthropic.messages.v1：Messages 只接受 Anthropic 原生 block type；不要提交 MiniMax 扩展的 video 或 mid_conv_system，它们会在解码阶段被拒绝。 原始契约：video and mid_conv_system content blocks are refused by the Messages decoder, which accepts only Anthropic’s own block types — MiniMax extends the wire form and Halro does not follow it there
   • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
   • minimax.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • kimi.anthropic.messages.v1：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
   • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
   • kimi.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   ↩

5. messages[].content 的组合级限制：

   • openai.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • minimax.anthropic.messages.v1：Messages 只接受 Anthropic 原生 block type；不要提交 MiniMax 扩展的 video 或 mid_conv_system，它们会在解码阶段被拒绝。 原始契约：video and mid_conv_system content blocks are refused by the Messages decoder, which accepts only Anthropic’s own block types — MiniMax extends the wire form and Halro does not follow it there
   • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
   • minimax.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • kimi.anthropic.messages.v1：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
   • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
   • kimi.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   ↩

6. messages[].content[].type=document 的组合级限制：

   • openai.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • minimax.anthropic.messages.v1：Messages 只接受 Anthropic 原生 block type；不要提交 MiniMax 扩展的 video 或 mid_conv_system，它们会在解码阶段被拒绝。 原始契约：video and mid_conv_system content blocks are refused by the Messages decoder, which accepts only Anthropic’s own block types — MiniMax extends the wire form and Halro does not follow it there
   • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
   • minimax.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • kimi.anthropic.messages.v1：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
   • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
   • kimi.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   ↩

7. messages[].content[].type=search_result 的组合级限制：

   • openai.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • minimax.anthropic.messages.v1：Messages 只接受 Anthropic 原生 block type；不要提交 MiniMax 扩展的 video 或 mid_conv_system，它们会在解码阶段被拒绝。 原始契约：video and mid_conv_system content blocks are refused by the Messages decoder, which accepts only Anthropic’s own block types — MiniMax extends the wire form and Halro does not follow it there
   • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
   • minimax.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • kimi.anthropic.messages.v1：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
   • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
   • kimi.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   ↩

8. stream 的组合级限制：

   • openai.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
   • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
   • minimax.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   • kimi.responses.v1：该 Profile 没有流式 primitive；stream=true 时 Halro 会在 Provider I/O 前排除该目标，请配置其他流式目标或改为非流式。 原始契约：this profile binds no stream primitive, so a streaming Messages request is routed away rather than refused by field
   ↩

9. stop_sequences 的组合级限制：

   • minimax.anthropic.messages.v1：MiniMax Messages 的原生和便携模式都不接受 top_k、stop_sequences 或 cache_control；提交前删除。 原始契约：top_k, stop_sequences and cache_control are rejected on the native path as well as the portable one, because MiniMax documents them as ignored rather than refused and a silently dropped boundary is billed as if it had been honored
   • kimi.anthropic.messages.v1：此面已实测执行 stop_sequences，不会像忽略该字段的 MiniMax 面一样拒绝它。 原始契约：stop_sequences is carried and was measured honored, so it is not refused the way MiniMax’s is
   • kimi.chat.v1：Kimi 已实测执行 stop_sequences，最多五项、每项 32 字节；超出则在 Provider I/O 前排除目标。 原始契约：stop_sequences is carried and was measured honored, within Kimi’s published bounds of five sequences of at most 32 bytes each; a request outside them is routed away before provider I/O
   ↩

10. temperature 的组合级限制：

    • kimi.anthropic.messages.v1：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected on both paths because Kimi pins each per model and answers any other value with an error
    • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
    • kimi.responses.v1：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected for the same reason as on the Kimi Chat profile: Kimi models neither member
    • kimi.code.anthropic.messages.v1：此面接受 temperature、top_k 并返回 200 却不能证明生效，因此 Halro 拒绝这些字段。 原始契约：temperature and top_k are refused here because this face accepts them and answers 200 without establishing that either was read, where the Chat face refuses them outright
    ↩

11. top_p 的组合级限制：

    • kimi.anthropic.messages.v1：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected on both paths because Kimi pins each per model and answers any other value with an error
    • kimi.chat.v1：Kimi Messages 请求必须省略 temperature 和 top_p；任一字段出现都会在 Provider I/O 前排除该 Profile。 原始契约：temperature and top_p are this endpoint’s own members and Kimi models neither: they are absent from its request schema and pinned per model, so a Messages request naming either is routed away before provider I/O
    • kimi.responses.v1：Kimi 对 temperature 和 top_p 使用模型固定值；请求中必须省略这两个字段，否则该 Profile 不可用。 原始契约：temperature and top_p are rejected for the same reason as on the Kimi Chat profile: Kimi models neither member
    ↩

12. top_k 的组合级限制：

    • minimax.anthropic.messages.v1：MiniMax Messages 的原生和便携模式都不接受 top_k、stop_sequences 或 cache_control；提交前删除。 原始契约：top_k, stop_sequences and cache_control are rejected on the native path as well as the portable one, because MiniMax documents them as ignored rather than refused and a silently dropped boundary is billed as if it had been honored
    • kimi.anthropic.messages.v1：Kimi Messages 的原生和便携模式都不接受 top_k 或 cache_control；提交前删除这两个字段。 原始契约：top_k and cache_control are rejected on the native path as well as the portable one: Kimi’s Messages schema has neither member, top_k answers 200 without anything establishing what it did, and prompt caching is automatic with no handle to mark
    • kimi.code.anthropic.messages.v1：此面接受 temperature、top_k 并返回 200 却不能证明生效，因此 Halro 拒绝这些字段。 原始契约：temperature and top_k are refused here because this face accepts them and answers 200 without establishing that either was read, where the Chat face refuses them outright
    ↩

13. tools 的组合级限制：

    • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    ↩

14. tools[].type=custom 的组合级限制：

    • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    ↩

15. tools[].type=bash_* 的组合级限制：

    • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    ↩

16. tools[].type=text_editor_* 的组合级限制：

    • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    ↩

17. tools[].type=memory_* 的组合级限制：

    • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    ↩

18. tools[].type=computer_* 的组合级限制：

    • bedrock.mantle.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    • bedrock.mantle.openai.responses.v1：不要组合 stream=true 与 tools；该 Profile 会在 Provider I/O 前拒绝这类请求。 原始契约：streaming requests with tools are rejected before provider I/O
    ↩

19. tool_choice 的组合级限制：

    • kimi.chat.v1：Kimi 与 reasoning 同用时不要指定命名工具；required 仅在 kimi-k3 可与 depth 同用，K2.x 会在预留后拒绝，生产 Route 应按模型能力拆分。 原始契约：tool_choice is value-dependent: a named function together with a depth is refused by every Kimi model and routed away before provider I/O, while required together with a depth is refused by the K2.x line alone — a per-model limit this declaration cannot express, refused after the reservation rather than costing kimi-k3 the request it serves
    ↩

20. thinking 的组合级限制：

    • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
    • kimi.chat.v1：该 Kimi Messages Profile 仅在未请求 reasoning 时把必填 max_tokens 映射到 max_completion_tokens；带 reasoning depth 的请求不可路由。 原始契约：max_tokens is carried as max_completion_tokens, which Kimi documents as the same bound under a newer name — but the two are the same quantity only while nothing is thinking, because Kimi’s single output bound counts reasoning, measured 2026-09-01. This endpoint requires max_tokens and means the answer alone, so a request that also asks for a depth is routed away rather than having this layer decide which of the two quantities the caller meant
    ↩

21. output_config 的组合级限制：

    • gemini.generate-content.text.v1beta：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O
    • bedrock.runtime.converse.text.v1：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O
    • bedrock.mantle.responses.v1：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry, so a request carrying it is routed away before provider I/O
    • bedrock.mantle.openai.responses.v1：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry, so a request carrying it is routed away before provider I/O
    • bedrock.mantle.anthropic.messages.v1：便携模式必须省略 output_config；原生模式才会原样转发，且仍受构建固定的 Mantle Beta 能力边界约束。 原始契约：output_config is unsupported in portable mode only: this profile shares the Anthropic wire form and could carry the member, but its Mantle Beta capability ceiling is fixed by the build and widening it is a separate contract review, so a portable request carrying effort or format is routed away before provider I/O while native mode forwards the member verbatim
    • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
    ↩

22. output_config.effort 的组合级限制：

    • gemini.generate-content.text.v1beta：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O
    • bedrock.runtime.converse.text.v1：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O
    • bedrock.mantle.responses.v1：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry, so a request carrying it is routed away before provider I/O
    • bedrock.mantle.openai.responses.v1：该无状态 Responses Profile 不传 output_config.effort；请求必须省略，否则目标会在 Provider I/O 前被排除。 原始契约：output_config.effort is this endpoint’s spelling of reasoning_effort, which the stateless Responses primitive does not carry, so a request carrying it is routed away before provider I/O
    • minimax.chat.v1：MiniMax 只能把任意 depth 映射为 thinking 开启，且不能与必填 max_tokens 同用；Messages 请求携带 effort 时该 Profile 会被排除。 原始契约：output_config.effort reaches MiniMax’s thinking switch, which has one on state and no depth ladder, so a depth is honored as on rather than preserved — but only on a request that does not also bound the answer, and this endpoint requires max_tokens, so an effort-bearing Messages request is routed away from this profile by field rather than translated
    ↩

23. output_config.format 的组合级限制：

    • gemini.generate-content.text.v1beta：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O
    • bedrock.runtime.converse.text.v1：该 Profile 不接受 output_config.effort 或 output_config.format 的任何值；必须省略这两个字段。 原始契约：output_config.effort and output_config.format are this endpoint’s spelling of reasoning_effort and response_format, which this profile declares unsupported at every value, so a request carrying either is routed away before provider I/O
    ↩