MaaS_DeepSeek
Model Overview
| item | value |
|---|---|
| Model ID (model) | deepseek-flash |
| context window | 1M (marked as "1M" on the origin server; the source document does not specify the exact token count) |
| Maximum Output | Up to 384K; the max_tokenscap for Chat Completions is 393216 |
| Default output limit | Non-reasoning mode: 8K; reasoning mode: 64K;when reasoning_effort= max, 128K |
| Max Chain-of-Thought | Source document not specified |
| input modality | Text, Image |
| output modality | Text |
| Default Thinking Mode | Enable (thinking.typedefaults to enabled, with the default effort set to high) |
| thinking gear | none / low / high / max; minimal→low, medium, xhigh→high, ultra→max |
| Image Support | Supported, only usermessages; JPEG / PNG / GIF / WebP; a single image is converted to a maximum of 1024 tokens |
| Context Caching | Supported, disk caching is enabled by default, no parameters required |
Request Protocol
https
Header
| Parameter Name | Type | Required | Description |
|---|---|---|---|
| Content-Type | string | Yes | application/json; multipart/form-data for file uploads |
| Authorization | string | Yes | Bearer {API_KEY} |
OpenAI Chat Completions
Request URL
POST https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions
Request Body Parameters
| Parameter Name | Secondary Parameters | Three-level parameters | CET-4 Parameters | Type | Required | Description |
|---|---|---|---|---|---|---|
| model | - | - | - | string | Yes | Fixed as MaaS_Deepseek_V4.1_Flash_20260910 (vendor's original model: deepseek-flash; optional origin server options: deepseek-flash, deepseek-v4-pro) |
| messages | - | - | - | array[json] | Yes | Conversation message list, with at least 1 entry |
| role | - | - | string | Yes | system / user / assistant / tool |
|
| content | - | - | string or array | Yes | system, toolare strings;usermessages can be an array of content blocks for image passing |
|
| type | - | string | Yes | Content block type: text/ image_url/ file |
||
| text | - | string | Conditions | type= textthe text content |
||
| image_url | - | object | Conditions | type= image_urlwhen in use |
||
| url | string | Yes | http (s)external link (max. 8192 characters) or base64 data URL |
|||
| detail | string | No | low (downsampled to 512×512) / high/ original/ auto;highis equivalent to original |
|||
| file_id | - | string | Conditions | type= file, the ID returned by the Files API (in the format of file-api-. ..) |
||
| file_data | - | string | Conditions | type= file, inline base64 data URL within the request, which is mutually exclusive with file_id |
||
| filename | - | string | No | Can be set when using file_data; not allowed to be used together with file_id |
||
| name | - | - | string | No | Participants can choose a name to distinguish participants with the same role | |
| tool_call_id | - | - | string | Conditions | role= toolis required, and the corresponding tool call ID shall be sent via postback |
|
| reasoning_content | - | - | string | Conditions | Chain of thought from the previous round in thinking mode; when the request carries tools, a complete postback is mandatory, otherwise a 400 will be returned |
|
| prefix | - | - | boolean | No | Prefix Continuation Switch (Beta, see "Chat Prefix Completion") | |
| thinking | - | - | - | object | No | Thinking Mode Switch:{"type":"enabled"\|"disabled"}, default enabled |
| reasoning_effort | - | - | - | string | No | none/ low/ high/ max, default: high;nonedisables reasoning |
| max_tokens | - | - | - | integer | No | Value ranges from 1 to 393216; when not set, it is 8K for non-thinking mode and 64K for thinking mode effort= max, and 128K in this case |
| response_format | - | - | - | object | No | {"type":"json_object"}Enable JSON output; default{"type":"text"} |
| stop | - | - | - | string or array | No | Up to 16 stop sequences |
| stream | - | - | - | boolean | No | Enable SSE streaming, ending with data: [DONE] |
| stream_options | - | - | - | object | No | Must be used together with stream: true, otherwise a 400 error will be returned |
| include_usage | - | - | boolean | No | When set to true, each chunk carries a usagefield (only the last non-empty one is valid); the last chunk always carries the full usage information regardless of the setting. |
|
| temperature | - | - | - | number | No | 0\~2, default value is 1; invalid in thinking mode (no error will be reported) |
| top_p | - | - | - | number | No | 0\~1, default is 1; in reasoning mode, values below 0.95 will be raised to 0.95; in non-reasoning mode, it is fixed at 1.0 and the input value is ignored |
| tools | - | - | - | array | No | Tool list, only supports function; tool names must be unique |
| type | - | - | string | Yes | fixed function |
|
| function | - | - | object | Yes | function definition | |
| name | - | string | Yes | Only a-z, A-Z, 0-9, _, -allowed, maximum 128 characters |
||
| description | - | string | No | for the model to determine when to invoke | ||
| parameters | - | object | No | JSON Schema; omitted indicates no parameters | ||
| strict | - | boolean | No | Default: false; Strict mode (Beta) requires the /betabase_url, and mandates that all objectshave additionalProperties= falsewith all fields marked as required |
||
| tool_choice | - | - | - | string or object | No | none/ auto/ required, or{"type":"function", "function":{"name":". .."}}; defaults to nonewhen no tools are present, defaults to autowhen tools are present;requiredwill return 400 with the specified tool in reasoning mode |
| logprobs | - | - | - | boolean | No | Whether to return the log probabilities of output tokens |
| top_logprobs | - | - | - | integer | No | 0\~20, required to simultaneously logprobs= true |
| user_id | - | - | - | string | No | Character set[a-zA-Z0-9\-_], maximum length 512; used for content security, KVCache isolation and scheduling isolation |
| frequency_penalty | - | - | - | number | No | Deprecated. Passing a value here has no effect. |
| presence_penalty | - | - | - | number | No | Deprecated. Passing a value here has no effect. |
Request Example
curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer {API_KEY}' \
-d '{
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Hello!"}
],
"thinking": {"type": "enabled"},
"reasoning_effort": "high",
"stream": false
}'
Response Example
{
"id": "chatcmpl-xxxxxxxxxxxxxxxx",
"object": "chat.completion",
"created": 1700000000,
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"system_fingerprint": "xxxxxxxx",
"choices": [
{
"index": 0,
"finish_reason": "stop",
"message": {
"role": "assistant",
"content": "Hello! How can I help you today?",
"reasoning_content": null,
"tool_calls": null
},
"logprobs": null
}
],
"usage": {
"prompt_tokens": 12,
"completion_tokens": 10,
"total_tokens": 22,
"prompt_tokens_details": {
"cached_tokens": 0,
"prompt_cache_hit_tokens": 0,
"prompt_cache_miss_tokens": 12
},
"completion_tokens_details": {
"reasoning_tokens": 0
}
}
}
finish_reasonpossible values: stop/ length/ content_filter/ tool_calls/ insufficient_system_resource/ aborted. In thinking mode, the chain of thought is located in message.reasoning_content, which is at the same level as content.
Example of streaming request
curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer {API_KEY}' \
-d '{
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"messages": [{"role": "user", "content": "Hello!"}],
"stream": true,
"stream_options": {"include_usage": true}
}'
```plain text data: {"id":"chatcmpl-xxx","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}],"usage":null}
data: {"id":"chatcmpl-xxx","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}],"usage":{"prompt_tokens":6,"completion_tokens":2,"total_tokens":8,"prompt_tokens_details":{"prompt_cache_hit_tokens":0,"prompt_cache_miss_tokens":6}}}
data: [DONE]
### SDK Calling Method
```python
from openai import OpenAI
client = OpenAI(
api_key="{API_KEY}",
base_url="https://genaiapi.cloudsway.net/v1/ai/{endpointPath}",
)
response = client.chat.completions.create(
model="MaaS_Deepseek_V4.1_Flash_20260910",
messages=[{"role": "user", "content": "Hello!"}],
reasoning_effort="high",
extra_body={"thinking": {"type": "enabled"}},
)
print(response.choices[0].message.content)
Note: When using the OpenAI SDK to pass thinking, it should be placed in extra_body;reasoning_effortis a top-level parameter
OpenAI Responses
Request URL
POST https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses
Request Body Parameters
| Parameter Name | Secondary Parameters | Three-level parameters | Type | Required | Description |
|---|---|---|---|---|---|
| model | - | - | string | Yes | Fixed as MaaS_Deepseek_V4.1_Flash_20260910 (vendor's original model: deepseek-flash) |
| input | - | - | string or array | Conditions | Pure strings are processed as individual usermessages, or as an array of input items; at least one of this field and instructionsmust be filled in. |
| type | - |
string | No | Input item type: message/ function_call/ function_call_output/ custom_tool_call/ custom_tool_call_output/ reasoning; other types are ignored |
|
| role | - | string | No | Role of themessage item: user/ assistant/ system/ developer (developeris regarded as user) |
|
| content | - | string or array | No | content string or content block array | |
| type | string | Yes | input_text / output_text / input_image |
||
| text | string | Conditions | Text content block | ||
| image_url | string | Conditions | input_imageis the http (s)link or base64 data URL of the image, and it is mutually exclusive with file_id. |
||
| file_id | string | Conditions | input_imagereferences the image from the Files API;file_idis present, detailis ignored |
||
| detail | string | No | low/ high/ original/ auto; lowdownsampled to 512×512 |
||
| output | - | string or array | No | function_call_output/ custom_tool_call_outputoutputs, which can be an array of content blocks (supporting input_image) |
|
| instructions | - | - | string | No | System-level instruction, inserted as the first system message |
| reasoning | - | - | object | No | Thinking Mode Configuration |
| effort | - | string | No | none/ low/ high/ max;nonedisables reasoning;minimal→ low, medium, xhigh→ high; model default (enabled) applies if not set |
|
| summary | - | string | No | The origin server accepts this field but does not generate a digest | |
| max_output_tokens | - | - | integer | No | Output limit, including visible output and inference tokens |
| stream | - | - | boolean | No | Semantic SSE, with termination events being response.completed/ response.incomplete/ response.failed, and no data: [DONE] |
| temperature | - | - | number | No | 0\~2, default value is 1; invalid in thinking mode |
| top_p | - | - | number | No | 0\~1, default value is 1; the lower limit is 0.95 in thinking mode, and fixed at 1.0 in non-thinking mode |
| text | - | - | object | No | Text Output Configuration |
| format | - | object | No | {"type":"text"} (default) / {"type":"json_object"} / {"type":"json_schema"} |
|
| type | string | No | text / json_object / json_schema |
||
| name | string | Conditions | type= json_schemais required |
||
| schema | object | Conditions | type= json_schemais required, and the output must conform to this JSON Schema |
||
| verbosity | - | string | No | The origin server accepts this field but does not activate it | |
| tools | - | - | array | No | Only functiontakes effect;customonly supports{"type":"custom", "name":"apply_patch"}; built-in tools (web_search, file_search, code_interpreter, etc.) are ignored |
| type | - | string | Yes | function |
|
| name | - | string | Yes | Non-empty, no more than 128 characters, matching ^[a-zA-Z0-9_-]+$, and globally unique |
|
| description | - | string | No | for the model to determine when to invoke | |
| parameters | - | object | No | JSON Schema; omitted indicates no parameters | |
| tool_choice | - | - | string or object | No | none / auto (default) / required, or {"type":"function","name":"..."} |
| top_logprobs | - | - | integer | No | 0\~20 |
| user | - | - | string | No | Character set[a-zA-Z0-9\-_], with a maximum length of 512; used for security audit, KVCache and scheduling isolation. |
| parallel_tool_calls | - | - | boolean | No | Ignored (parallel tool calls are always enabled) |
| max_tool_calls | - | - | integer | No | Ignored |
| previous_response_id | - | - | string | No | Not supported (stateless interface) |
| conversation | - | - | - | No | Not supported |
| store | - | - | - | No | Not supported, the response is always store: false |
| background | - | - | - | No | Not supported |
| metadata | - | - | - | No | Not supported |
| include | - | - | - | No | Not supported |
| prompt | - | - | - | No | Not supported |
| truncation | - | - | - | No | Not supported; returns 400 when exceeding context limit |
| service_tier | - | - | - | No | Not supported |
| safety_identifier | - | - | - | No | Not supported |
| prompt_cache_key / prompt_cache_retention | - | - | - | No | Not supported; cache is managed automatically |
| context_management | - | - | - | No | Not supported |
| stream_options | - | - | - | No | Not supported |
Unsupported parameters are silently ignored without throwing errors. File input (of the filetype) is not supported; images are only allowed in user/ developermessages and the function_call_output/ custom_tool_call_output outputfield. If they appear in system/ assistantmessages, a 400 error will be returned.
Request Example
curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer {API_KEY}' \
-d '{
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"instructions": "You are a helpful assistant.",
"input": "Hi, how are you?",
"reasoning": {"effort": "high"}
}'
Response Example
{
"id": "resp-xxxxxxxxxxxxxxxx",
"object": "response",
"created_at": 1700000000,
"status": "completed",
"error": null,
"incomplete_details": null,
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"output": [
{
"type": "reasoning",
"id": "rs-xxx",
"status": "completed",
"content": [{"type": "reasoning_text", "text": "..."}]
},
{
"type": "message",
"id": "msg-xxx",
"status": "completed",
"role": "assistant",
"content": [{"type": "output_text", "text": "I am doing well, thanks!"}]
}
],
"usage": {
"input_tokens": 12,
"input_tokens_details": {"cached_tokens": 0},
"output_tokens": 20,
"output_tokens_details": {"reasoning_tokens": 8},
"total_tokens": 32
}
}
status values: in_progress / completed / incomplete / failed; incomplete_details.reason values: max_output_tokens / content_filter.
Example of streaming request
curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpoint}/responses' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer {API_KEY}' \
-d '{
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"input": "Hi, how are you?",
"stream": true
}'
(Event stream: response.created → response.in_progress → response.output_item.added/done → response.content_part.added/done → response.reasoning_text.delta/done → response.output_text.delta/done → response.function_call_arguments.delta/done → response.custom_tool_call_input.delta/done → terminates with response.completed / response.incomplete / response.failed, and each event carries a sequence_number.)
SDK Calling Method
from openai import OpenAI
client = OpenAI(
api_key="{API_KEY}",
base_url="https://genaiapi.cloudsway.net/v1/ai/{endpoint}",
)
response = client.responses.create(
model="MaaS_Deepseek_V4.1_Flash_20260910",
instructions="You are a helpful assistant.",
input="Hi, how are you?",
reasoning={"effort": "high"},
)
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 |
|---|---|---|---|
| x-api-key | string | Yes | origin server marked as Fully Supported (Anthropic protocol authentication header) |
| anthropic-version | string | No | Origin server is ignored; configuration passed will not take effect |
| anthropic-beta | string | Conditions | /messagesis ignored; required for Anthropic-compatible endpoints of Files files-api-2025-04-14 |
Request Body Parameters
| Parameter Name | Secondary Parameters | Three-level parameters | CET-4 Parameters | Type | Required | Description |
|---|---|---|---|---|---|---|
| model | - | - | - | string | Yes | Fixed as MaaS_Deepseek_V4.1_Flash_20260910 (vendor's original model: deepseek-flash) |
| max_tokens | - | - | - | integer | Yes | Origin server marked as Fully Supported; mandatory under this protocol |
| messages | - | - | - | array[json] | Yes | Anthropic message format, which differs from the messagesstructure of Chat Completions |
| role | - | - | string | Yes | user / assistant |
|
| content | - | - | string or array | Yes | Array of strings or content blocks | |
| type | - | string | Yes | text / image / thinking / tool_use / tool_result / server_tool_use / web_search_tool_result are supported; document, search_result, redacted_thinking, code_execution_tool_result, mcp_tool_use, mcp_tool_result, container_upload are not supported |
||
| text | - | string | Conditions | type= textfor text (cache_control, citationsare ignored) |
||
| source | - | object | Conditions | type= imageImage source at the time |
||
| type | string | Yes | base64 / url / file |
|||
| media_type | string | Conditions | Required when base64: image/jpeg / image/png / image/gif / image/webp |
|||
| data | string | Conditions | base64image data |
|||
| url | string | Conditions | urlexternal link, with a maximum length of 8192 characters |
|||
| file_id | string | Conditions | fileWhen referencing images via the Files API, the following header must be included: anthropic-beta: files-api-2025-04-14 |
|||
| id / name / input / tool_use_id / content / is_error | - | - | Conditions | tool_use / tool_result fields: id, input, name, tool_use_id, content are supported; cache_control, is_error are ignored |
||
| system | - | - | - | string | No | origin server labeling Fully Supported |
| stream | - | - | - | boolean | No | Origin server annotation: Fully Supported (not specified in the source document of the event list) |
| stop_sequences | - | - | - | array | No | origin server labeling Fully Supported |
| temperature | - | - | - | number | No | The origin server is marked as Fully Supported, with the range [0.0 \~ 2.0] |
| top_p | - | - | - | number | No | Only the thinking mode is enabled, with a lower limit of 0.95; the non-thinking mode is fixed at 1.0 |
| top_k | - | - | - | integer | No | origin server ignored |
| thinking | - | - | - | object | No | Supported;budget_tokensis ignored |
| output_config | - | - | - | object | No | Only effortis supported |
| effort | - | - | string | No | low/ high/ max (the name of the thinking tier field under the Anthropic protocol) |
|
| metadata | - | - | - | object | No | Only user_idis supported, others are ignored |
| user_id | - | - | string | No | for security audit, KVCache and scheduling isolation | |
| tools | - | - | - | array | No | Tool Definition |
| name | - | - | string | Yes | origin server marked as Fully Supported | |
| input_schema | - | - | object | Yes | origin server marked as Fully Supported | |
| description | - | - | string | No | origin server marked as Fully Supported | |
| cache_control | - | - | - | No | Ignored | |
| tool_choice | - | - | - | object | No | nonesupported;auto/ any/ toolsupported (disable_parallel_tool_useis ignored) |
| container | - | - | - | - | No | Ignored |
| mcp_servers | - | - | - | - | No | Ignored |
| service_tier | - | - | - | - | No | Ignored |
Note: The origin server maps Claude model names as follows — claude-opus*→ deepseek-v4-pro, claude-haiku*, claude-sonnet*→ deepseek-flash; unrecognized model names passed in will automatically fall back to the underlying Flash model. Whether the MaaS side will inherit this mapping is subject to the platform configuration.This protocol supports inserting tool call messages and systemmessages in the middle of a conversation (the Chat Completions protocol does not support inserting tool calls).
Request Example
curl -X POST 'https://genaiapi.cloudsway.net/{endpoint}/v1/messages' \
-H 'Content-Type: application/json' \
-H 'x-api-key: {API_KEY}' \
-d '{
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"max_tokens": 1000,
"system": "You are a helpful assistant.",
"messages": [
{"role": "user", "content": [{"type": "text", "text": "Hi, how are you?"}]}
],
"output_config": {"effort": "high"}
}'
Response Example
{
"id": "msg_xxxxxxxxxxxxxxxx",
"type": "message",
"role": "assistant",
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"content": [
{"type": "text", "text": "I am doing well, thanks!"}
],
"stop_reason": "end_turn",
"stop_sequence": null,
"usage": {"input_tokens": 12, "output_tokens": 20}
}
(contentand usageshall be subject to the specific field names returned by the MaaS endpoint; the origin server compatibility table only provides the field support status, not the complete list of response fields)
Example of streaming request
curl -X POST 'https://genaiapi.cloudsway.net/{endpoint}/v1/messages' \
-H 'Content-Type: application/json' \
-H 'x-api-key: {API_KEY}' \
-d '{
"model": "MaaS_Deepseek_V4.1_Flash_20260910",
"max_tokens": 1000,
"messages": [{"role": "user", "content": "Hi, how are you?"}],
"stream": true
}'
(origin server marked streamFully Supported, but the list of event types has not been published — "not specified in the source document")
SDK Calling Method
import anthropic
client = anthropic.Anthropic(
api_key="{API_KEY}",
base_url="https://genaiapi.cloudsway.net/{endpoint}/v1",
)
message = client.messages.create(
model="MaaS_Deepseek_V4.1_Flash_20260910",
max_tokens=1000,
system="You are a helpful assistant.",
messages=[
{"role": "user", "content": [{"type": "text", "text": "Hi, how are you?"}]}
],
)
print(message.content)