MaaS_Meta_muse_wonder_1.0_20261012
Model Overview
| item | value |
|---|---|
| Model ID (model) | muse-wonder-1.0 |
| context window | 1,048,576 tokens |
| Input Modality | Text, Image, Video (not supported yet), Audio (not supported yet), PDF (not supported yet) |
| output modality | Text |
| Default Thinking Mode | Always enabled. Inference will still be performed even when no gear is specified, and the gear will be determined by the model. none/ thinking.type= disabledReturns HTTP 400 |
| Inference Gear | minimal、low、medium、high、xhigh、max。max 不会切换到 Muse Wonder Full |
Request Protocol
https
Header
| Parameter Name | Type | Required | Description |
|---|---|---|---|
| Content-Type | string | Yes | application/json |
| Authorization | string | Yes | Bearer ${YOUR_AK} |
OpenAI Chat Completions
Request URL
POST https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions
Request Body Parameters
| Parameter Name | Secondary Parameters | Three-level parameter | CET-4 Parameters | Type | Required | Description |
|---|---|---|---|---|---|---|
| model | - | - | - | string | Yes | Fixed as muse-wonder-1.0 (original manufacturer model: MaaS_Meta_muse_wonder_1.0_20261012) |
| messages | - | - | - | array[json] | Yes | Conversations read in sequence. At least 1 entry. Roles: developer (highest priority), system (same priority as developer, recommended to use developer instead), user, assistant, tool. No cross-request memory retained. When assistantcontains tool_calls, a matching tool_call_id toolmessage must be supplemented for each call before the next round; otherwise, an HTTP 400 error will be triggered. This endpoint does not retain reasoning across rounds. |
| role | - | - | string | Yes | developer / system / user / assistant / tool |
|
| content | - | - | string 或 array | Conditions | assistantcan be empty when carrying tool_calls, and is required for other roles. usercontent can be plain text, or a text/ image_url/ input_audio/ file/ video_urlcontent block. |
|
| content | type | - | string | Conditions | Content Block Type | |
| content | text | - | string | Conditions | type= texttext when applicable |
|
| content | image_url | url | string | Conditions | type=image_url。公网 URL 或 base64 data URL |
|
| content | image_url | detail | string | No | auto (default), low, high |
|
| content | input_audio | data | string | Conditions | type=input_audio。base64 音频 |
|
| content | input_audio | format | string | Conditions | wavor mp3 |
|
| content | audio_url | - | string | No | Audio URL or data URL | |
| content | file | file_data | string | No | type= file. inline base64 |
|
| content | file | filename | string | No | inline file name | |
| content | video_url | url | string | Conditions | type=video_url。视频 URL 或 base64 data URI |
|
| content | video_url | fps | number | No | Frame extraction frame rate. The source document does not specify the value range | |
| name | - | - | string | No | Distinguish participants with the same role | |
| tool_calls | - | - | array | No | Function call of the playback assistant. The element contains id, type= function, function.name, function.arguments (a JSON string that must be validated before the call) |
|
| tool_call_id | - | - | string | Conditions | role= toolis required, corresponding to the call ID of the response |
|
| reasoning_content | - | - | string | No | The requesting side can perform playback; this field will be cleared in the response from the external caller and cannot be used to resume inference. reasoningis its alias. If both exist, reasoning_contentshall prevail. |
|
| temperature | - | - | - | number | No | 0~2, default value is 1. Choose either this parameter or top_p, and it is officially recommended to keep the default setting |
| top_p | - | - | - | number | No | Defaults to 1, must satisfy 0 < top_p ≤ 1. If 0is passed, HTTP 400 will be returned |
| max_completion_tokens | - | - | - | integer | No | Indicates the combined upper limit of output and inference tokens. It shares the 1,048,576 context window with the input; an HTTP 400 will be returned if the model's upper limit is exceeded. The deprecated alias max_tokensis still accepted |
| stream | - | - | - | boolean | No | Defaults to false. When set to true, results are returned via SSE, with the end marker being data: [DONE]. Non-streaming requests that take too long may result in an HTTP 504 error, while streaming requests are not subject to this time limit. |
| stream_options | include_usage | - | - | boolean | No | When set to true, the usage only appears in the last chunk; an interrupted stream may not have this chunk |
| reasoning_effort | - | - | - | string | No | MaaS_Meta_muse_wonder_1. 0_20261012Available minimal, low, medium, high, xhigh, max. Maxis longer than xhigh, do not switch Muse Wonder Full. Nonereturns HTTP 400. The model still inferences when not transmitted. The inference token is counted in the max_completion_tokensand output billing, and the response does not return the thought chain body |
| response_format | type | - | - | string | No | text, json_object, json_schema. Structured output gives priority to json_schema;json_objectonly guarantees valid JSON, and it is required that the prompt calls for outputting JSON |
| json_schema | name | - | string | Conditions | type= json_schemais required. The maximum length is 64 characters, which can only be letters, digits, underscores or hyphens. |
|
| json_schema | schema | - | object | No | JSON Schema | |
| json_schema | strict | - | boolean | No | Defaults to false. When set to true, strict schema matching is required, and only a subset of JSON Schema is supported | |
| json_schema | description | - | string | No | Format Usage Instructions | |
| tools | - | - | - | array[json] | No | 仅 type=function |
| type | - | - | string | Yes | fixed function |
|
| function | name | - | string | Yes | Function name. Consists of letters, numbers, underscores, hyphens, with at most one dot | |
| function | description | - | string | No | for the model to decide whether to call | |
| function | parameters | - | object | No | JSON Schema. Omission indicates no parameters | |
| function | strict | - | boolean | No | Defaults to false | |
| tool_choice | - | - | - | string | No | Only auto. Defaults to toolswhen auto is passed. |
| parallel_tool_calls | - | - | - | boolean | No | Whether to allow parallel function calls in the same round |
| n | - | - | - | integer | No | Only 1. Return HTTP 400 if the value is greater than 1 |
| prompt_cache_retention | - | - | - | string | No | in_memoryor 24h. The actual retention period is determined by the server |
| prompt_cache_key | - | - | - | string | No | Group similar requests to improve cache hits, replacing the cache usage of user |
| safety_identifier |
- | - | - | string | No | A stable identifier for the end user, with a maximum length of 64 characters. It is recommended to pass a hash value instead of the original username or email address. When this field coexists with user, this field shall prevail |
| user | - | - | - | string | No | Deprecated. For abuse detection, use safety_identifierinstead; for caching, use prompt_cache_key |
| service_tier (not supported yet) | - | - | - | string | No | schema enumeration auto, default, flex, scale, priority. The source document does not specify the differences between the various tiers of this model. |
| metadata (not supported yet) | - | - | - | object | No | Source document unexpanded key |
| store (not supported yet) |
- |
- | - | boolean |
No |
Defaults to false. Determines whether to save outputs for distillation or evaluation; supports text and images, and images larger than 8MB will be discarded. |
Request Example
curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer ${YOUR_AK}' \
-d '{
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"reasoning_effort": "high",
"max_completion_tokens": 4096,
"messages": [
{"role": "user", "content": "用三句话说明 RAG 与长上下文的取舍。"}
]
}'
Response Example
{
"id": "chatcmpl-abc123",
"object": "chat.completion",
"created": 1714502400,
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "RAG 在检索后再生成;长上下文把材料直接放进窗口。"
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 25,
"completion_tokens": 19,
"total_tokens": 44,
"completion_tokens_details": {
"reasoning_tokens": 0
}
}
}
finish_reasonValid values: stop, length, tool_calls, content_filter, function_call. The main text is located at choices[0]. message.content. The reasoning_contentin the external response is empty.
Example of streaming request
curl -N -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer ${YOUR_AK}' \
-H 'Accept: text/event-stream' \
-d '{
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"messages": [
{"role": "user", "content": "解释 Python 垃圾回收。"}
],
"stream": true
}'
The delta is in choices[0]. delta.content. The last chunk before the end carries finish_reason, and the stream terminates with data: [DONE]. Inference occurs before the visible text, so there will be a waiting period for the first packet.
SDK Calling Method
from openai import OpenAI
client = OpenAI(
base_url="https://genaiapi.cloudsway.net/v1/ai/{endpointPath}",
api_key="${YOUR_AK}",
)
response = client.chat.completions.create(
model="MaaS_Meta_muse_wonder_1.0_20261012",
reasoning_effort="high",
messages=[{"role": "user", "content": "用三句话说明 RAG 与长上下文的取舍。"}],
)
print(response.choices[0].message.content)
OpenAI Responses
Request URL
POST https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses
Request Body Parameters
| Parameter Name | Secondary Parameters | Three-level parameter | Type | Required | Description |
|---|---|---|---|---|---|
| model | - | - | string | Yes | Fixed as MaaS_Meta_muse_wonder_1.0_20261012 (Original model by vendor: MaaS_Meta_muse_wonder_1.0_20261012) |
| input | - | - | string 或 array | Yes | A string, or an array of typed items. The message roles include developer, system, user, and assistant. User content blocks include input_text, input_image, input_file, and input_video; the assistant's text for playback uses output_text. There are also reasoning, function_call, and function_call_output |
| role | - | string | Conditions | Message item: user/ assistant/ system/ developer. developertakes precedence over system |
|
| content | type | string | Conditions | input_text、input_image、input_file、input_video,回放时为 output_text |
|
| content | text | string | Conditions | input_text/ output_texttext. input_text.texthas a maximum length schema of 10485760 |
|
| content | image_url | string | No | input_image。完整 URL 或 base64 data URL |
|
| content | detail | string | No | Image high, low, auto (default), original |
|
| phase | - | string | No | Assistant-only messages. commentaryrefers to the intermediate text before a tool call;final_answeris only accepted on the input side, and responses will not be tagged with this value. Intermediate text before a tool call must be marked as commentary, otherwise an HTTP 400 error will be returned. |
|
| type | - | string | Conditions | item 类型,如 message、reasoning、function_call、function_call_output |
|
| encrypted_content | - | string | Conditions | type= reasoningstateless replay ciphertext. Must be retained as a whole. Cannot be used in the same request as previous_response_id. |
|
| instructions | - | - | string | No | Insert developer instructions at the top of the context. When using previous_response_id, the instructions from the previous round will not be automatically carried over |
| max_output_tokens | - | - | integer | No | Specifies the maximum total number of output and inference tokens. The minimum value is 16. If the value exceeds the model's upper limit or the remaining context is insufficient, an HTTP 400 error will be returned. |
| temperature | - | - | number | No | 0~2, default value is 1. Choose either this parameter or top_p |
| top_p | - | - | number | No | Default value: 1. Must be greater than 0, ≤ 1;If ≤ 0, HTTP 400 is returned |
| reasoning | effort | - | string | No | MaaS_Meta_muse_wonder_1.0_20261012: minimal, low, medium, high, xhigh, max. nonereturns HTTP 400. maxdoes not switch to Muse Wonder Full; the model still performs inference when this parameter is not passed. The Responses schema will be marked as mediumby default, and the dedicated inference section specifies that the gear will be determined by the model if this parameter is omitted. |
| summary | - | string | No | This agreement only. auto (approximately one sentence), concise, detailed. There is no guarantee that there will be a summary for each round; when there is none, the summarywill be[]. The deprecated alias generate_summarywill not actually request a summary |
|
| stream | - | - | boolean | No | Defaults to false. SSE. Cannot be true at the same time as background. Streaming responses come with usage by default, and stream_options.include_usagehas no effect on this protocol. |
| stream_options | include_obfuscation | - | boolean | No | Only set when stream= true. Enabled by default, it adds random obfuscationcharacters to the delta; you may set it to false for trusted networks. |
| background | - | - | boolean | No | Defaults to false. Executes asynchronously and returns immediately with status: queued. Afterwards, you can polling by id, retrieve results via streaming or cancel the task. It cannot be enabled together with streamat the same time |
| previous_response_id | - | - | string | No | Continue writing from the stored response, only pass the newly added input. It cannot be used in the same request as include: ["reasoning.encrypted_content"], nor can it be used together with conversation |
| store | - | - | boolean | No | Defaults to true. Only when set to true can previous_response_idbe used for continuation. For stateless replay, it is recommended to set this to false and request encrypted inference |
| include | - | - | array[string] | No | This model can be used with reasoning.encrypted_content. message.output_text.logprobsreturns HTTP 400. Unknown values return HTTP 400 |
| tools | - | - | array[json] | No | function tool type= function;built-in internet access type= web_search (or web_search_2025_08_26) |
| type | - | string | Yes | function 或 web_search |
|
| name | - | string | Conditions | functionRequired |
|
| description | - | string | No | Function Description | |
| parameters | - | object | Conditions | function's JSON Schema. The schema marks this field as required |
|
| strict | - | boolean | No | Whether function parameters strictly match the schema. Defaults to true on the Responses side | |
| defer_loading | - | boolean | No | Delayed loading when set to true, discovered on demand via tool search | |
| filters | allowed_domains | array[string] | No | web_searchAllowed domains. If omitted, no domain-based restriction will be applied |
|
| user_location | - | object | No | Approximate location for web search | |
| search_context_size | - | string | No | 低、 中(默认)、 高 |
|
| tool_choice | - | - | string | No | schema only auto: the model independently decides whether to invoke a tool |
| parallel_tool_calls | - | - | boolean | No | Defaults to true. Specifies whether multiple tool calls are allowed in the same round. |
| max_tool_calls | - | - | integer | No | The maximum number of built-in tool calls, accumulated across all built-in tools. The minimum value is 1. Any excess calls will be ignored. |
| text | format | type | string | No | text、json_object 或 json_schema |
| truncation | - | - | string | No | Only disabled (default) is valid. autoreturns HTTP 400. Exceeding the context limit directly results in a failure, and the server does not truncate the content. |
| prompt_cache_retention | - | - | string | No | in_memory 或 24h |
| prompt_cache_key | - | - | string | No | Grouping keys for improving cache hit rate |
| safety_identifier | - | - | string | No | End-user identifier with a maximum length of 64 characters, hashing is recommended |
| user | - | - | string | No | Deprecated |
| service_tier | - | - | string | No | schema enumeration auto, default, flex, scale, priority. The source document does not specify the differences between the various tiers of this model. |
| metadata | - | - | object | No | Request metadata |
| top_logprobs | - | - | integer | No | 0~20. Depends on logprobs; requesting logprobs for this model will result in HTTP 400 |
| client_metadata | - | - | object | No | Compatibility field, ignored after acceptance |
| prompt | id | - | string | No | Stored prompt template ID. Ignored upon overall acceptance |
| context_management | - | - | array | No | Context management configuration. The source document has not expanded the items available for this model |
Request Example
curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer ${YOUR_AK}' \
-d '{
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"reasoning": {"effort": "high"},
"max_output_tokens": 4096,
"input": "用三句话说明 RAG 与长上下文的取舍。"
}'
Response Example
{
"id": "resp_abc123",
"object": "response",
"created_at": 1714502400,
"status": "completed",
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"output": [
{
"id": "msg_abc123",
"type": "message",
"role": "assistant",
"status": "completed",
"content": [
{
"type": "output_text",
"text": "RAG 先检索再生成;长上下文把材料直接放进窗口。",
"annotations": []
}
]
}
],
"usage": {
"input_tokens": 69,
"output_tokens": 163,
"total_tokens": 232,
"output_tokens_details": {"reasoning_tokens": 0},
"input_tokens_details": {"cached_tokens": 0}
}
}
When continuing the generation, use the returned idas the previous_response_idfor the next request. For stateless multi-turn conversations, include include: ["reasoning.encrypted_content"]in the request, and in the next turn, put the reasoningitem back into the inputas-is. The idof the reasoning item is required in the output; when replaying as input, the idcan be omitted, but encrypted_contentmust be included.
Example of streaming request
curl -N -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer ${YOUR_AK}' \
-d '{
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"input": "解释神经网络如何学习。",
"stream": true
}'
The visible text delta event is response.output_text.delta, with the field being delta. If reasoning.summaryis requested, the summary events are response.reasoning_summary_text.deltaand response.reasoning_summary_text.done. The original chain of thought is not streamed in plaintext.
SDK Calling Method
from openai import OpenAI
client = OpenAI(
base_url="https://genaiapi.cloudsway.net/v1/ai/{endpoint}",
api_key="${YOUR_AK}",
)
response = client.responses.create(
model="MaaS_Meta_muse_wonder_1.0_20261012",
reasoning={"effort": "max"},
input="用三句话说明 RAG 与长上下文的取舍。",
)
print(response.output_text)
Anthropic Messages
Request URL
POST https://genaiapi.cloudsway.net/{endpoint}/v1/messages
Header (Additional to This Agreement)
| Parameter Name | Type | Required | Description |
|---|---|---|---|
| anthropic-version | string | No | SDK compatibility header. The source document does not provide the required version number |
| x-api-key | string | No | The API key header used by the Anthropic SDK. Also accepts Authorization: Bearer |
Request Body Parameters
| Parameter Name | Secondary Parameters | Three-level parameter | CET-4 Parameters | Type | Required | Description |
|---|---|---|---|---|---|---|
| model | - | - | - | string | Yes | Fixed as MaaS_Meta_muse_wonder_1.0_20261012 (original manufacturer model: MaaS_Meta_muse_wonder_1.0_20261012). minLength 1 |
| max_tokens | - | - | - | integer | Yes | Maximum token generation limit, minimum value is 1. This maps to the underlying Responses' max_output_tokens, which includes inference tokens |
| messages | - | - | - | array[json] | Yes | At least one entry. contentis a string or an array of content blocks. The adaptation layer has no server-side session, and the history is replayed by the Client |
| role | - | - | string | Yes | The available schemas are user, assistant, system, and developer. The protocol specifies that the conversation is primarily structured around alternating turns between the user and the assistant, with system instructions placed at the top-level system |
|
| content | - | - | string 或 array | Yes | User blocks: text, image, video, document, tool_result, search_result. Assistant replay: text, tool_use, server_tool_use, thinking, redacted_thinking. Unsupported blocks return invalid_request_error |
|
| content | type | - | string | Conditions | Block Type | |
| content | text | - | string | Conditions | type= textor replay thinkingthe summary text |
|
| content | source | type | string | Conditions | Image, video and document sources: base64, url, file; documents can also be text, content |
|
| content | source | media_type | string | Conditions | base64 时的 MIME,如 image/png、application/pdf |
|
| content | source | data | string | Conditions | base64 bytes | |
| content | source | url | string | Conditions | type=url |
|
| content | fps | - | number | No | videoFrame extraction frame rate. The source document does not specify the value range |
|
| content | tool_use_id | - | string | Conditions | tool_resultRequired |
|
| content | id | - | string | Conditions | tool_use / server_tool_use 必填 |
|
| content | name | - | string | Conditions | tool_useis the function name. server_tool_usecan only be web_searchor tool_search |
|
| content | input | - | object | No | Tool parameters for playback | |
| content | data | - | string | No | redacted_thinkingmust be replayed as-is for its encrypted inference. The old field name redacted_thinkingis read when datais missing |
|
| system | - | - | - | string 或 array | No | A behavior instruction, equivalent to a developer message. This array only applies to type= text. All other block types will return HTTP 400 |
| type | - | - | string | Conditions | Fixed text |
|
| text | - | - | string | Conditions | System prompt text | |
| temperature | - | - | - | number | No | Limited to the range 0 to 1 for Anthropic. The official recommendation is to keep the default value of 1 for most workloads, and it should be used as an alternative to top_p (choose one of the two) |
| top_p | - | - | - | number | No | Passthrough. The schema ranges from 0 to 1 |
| stream | - | - | - | boolean | No | 默认 false。为 true 时按 Anthropic SSE:message_start、content_block_start、content_block_delta、content_block_stop、message_delta、message_stop |
| thinking | type | - | - | string | No | adaptive: Default mode, and output the summary. enabled: Reserved for compatibility, budget_tokensis not mapped to effort. disabled: This model returns HTTP 400. |
| budget_tokens | - | - | integer | Conditions | type= enabledis required. The minimum value is 1024 and must be less than max_tokens |
|
| display | - | - | string | No | summarized (default, returns a thinking summary) or omitted (returns only encrypted reasoning) |
|
| output_config | effort | - | - | string | No | low, medium, high, xhigh, max. maxindicates the highest gear currently available. The protocol specifies that these values will be transparently transmitted. This protocol schema does not include minimalnor none |
| format | type | - | string | Conditions | Fixed json_schema |
|
| format | schema | - | object | Conditions | JSON Schema to be complied with | |
| tools | - | - | - | array[json] | No | customfunction, web_search, tool_search. Unsupported hosted tools will return invalid_request_error |
| type | - | - | string | Conditions | Custom tools can be omitted or set to custom. Internet access tools: web_search, web_search_20250305, web_search_20250305_15. Tool discovery: tool_search, tool_search_20250305 |
|
| name | - | - | string | Conditions | Custom Tool Required | |
| description | - | - | string | No | Tool Description | |
| input_schema | - | - | object | No | JSON Schema for custom tool input parameters | |
| strict | - | - | boolean | No | Strict schema matching required | |
| defer_loading | - | - | boolean | No | When set to true, tool_searchloads the schema on demand |
|
| user_location | - | - | object | No | web_searchapproximate location: type= approximate, as well as city, region, country, and timezone |
|
| tool_choice | type | - | - | string | Conditions | auto, any (must be invoked), none (discard the tool for this round). If a named tool returns HTTP 400, namewill be ignored |
| disable_parallel_tool_use | - | - | boolean | No | When set to true, parallel tool calls in the same round are prohibited | |
| metadata | user_id | - | - | string | No | Identify the end user and write it to the safety identifier. Values exceeding 64 bytes will be hashed stably. Other string values are transparently passed through to Responses metadata |
| service_tier | - | - | - | string | No | Only auto, standard_onlyare supported. HTTP 400 will be returned for other values. |
| cache_control | type | - | - | string | No | Indicates a cache breakpoint, only ephemeral |
| ttl | - | - | string | No | Only 5m |
Request Example
curl -X POST 'https://genaiapi.cloudsway.net/{endpoint}/v1/messages' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer ${YOUR_AK}' \
-d '{
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"max_tokens": 1024,
"thinking": {"type": "adaptive"},
"output_config": {"effort": "high"},
"messages": [
{"role": "user", "content": "用三句话说明 RAG 与长上下文的取舍。"}
]
}'
Response Example
{
"id": "msg_abc123",
"type": "message",
"role": "assistant",
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"content": [
{"type": "text", "text": "RAG 先检索再生成;长上下文把材料直接放进窗口。"}
],
"stop_reason": "end_turn",
"stop_sequence": null,
"usage": {
"input_tokens": 14,
"output_tokens": 8,
"cache_read_input_tokens": 0,
"output_tokens_details": {"thinking_tokens": 0}
}
}
stop_reason: end_turn, max_tokens, tool_use, refusal. The schema also lists stop_sequence, pause_turn. Text is located at content[]. text; reasoning summary is thinking, encrypted reasoning is redacted_thinking; function calls are tool_use;web_search / tool_search 为 server_tool_use。
Example of streaming request
curl -N -X POST 'https://genaiapi.cloudsway.net/{endpoint}/v1/messages' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer ${YOUR_AK}' \
-d '{
"model": "MaaS_Meta_muse_wonder_1.0_20261012",
"max_tokens": 1024,
"stream": true,
"messages": [
{"role": "user", "content": "法国的首都是哪里?"}
]
}'
Content-Type: text/event-stream。事件顺序为 message_start、content_block_start、content_block_delta、content_block_stop、message_delta、message_stop。错误事件名为 error。
SDK Calling Method
from anthropic import Anthropic
client = Anthropic(
base_url="https://genaiapi.cloudsway.net/{endpoint}",
auth_token="${YOUR_AK}",
)
message = client.messages.create(
model="MaaS_Meta_muse_wonder_1.0_20261012",
max_tokens=1024,
thinking={"type": "adaptive"},
output_config={"effort": "high"},
messages=[{"role": "user", "content": "用三句话说明 RAG 与长上下文的取舍。"}],
)
print(message.content[0].text)