Responses - CometAPI Documentation

POST

responses

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cometapi.com/v1",
    api_key="<COMETAPI_KEY>",
)

response = client.responses.create(
    model="gpt-5.4",
    input="Tell me a three sentence bedtime story about a unicorn.",
)

print(response.output_text)

{
  "id": "resp_0a153ae8201f73bc0069a7e8044cc481",
  "object": "response",
  "created_at": 1772611588,
  "status": "completed",
  "background": false,
  "completed_at": 1772611589,
  "error": null,
  "incomplete_details": null,
  "instructions": null,
  "max_output_tokens": null,
  "model": "gpt-4.1-nano",
  "output": [
    {
      "id": "msg_0a153ae8201f73bc0069a7e8049a8881",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "type": "output_text",
          "annotations": [],
          "text": "Four."
        }
      ],
      "role": "assistant"
    }
  ],
  "parallel_tool_calls": true,
  "previous_response_id": null,
  "prompt_cache_key": null,
  "prompt_cache_retention": null,
  "reasoning": {
    "effort": null,
    "summary": null
  },
  "safety_identifier": null,
  "service_tier": "auto",
  "store": true,
  "temperature": 1,
  "text": {
    "format": {
      "type": "text"
    },
    "verbosity": "medium"
  },
  "tool_choice": "auto",
  "tools": [],
  "top_p": 1,
  "truncation": "disabled",
  "usage": {
    "input_tokens": 19,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 9,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 28
  },
  "user": null,
  "metadata": {}
}

Die Responses API erweitert Chat Completions um zustandsbehaftete Konversationen, integrierte Tools, multimodale Dateieingaben und Kontrolle über das Reasoning. Sie ist der empfohlene Endpunkt für OpenAI o-series Reasoning-Modelle, die GPT-5-Serie und Codex-Modelle.

Verschiedene Modellanbieter unterstützen unterschiedliche Request-Parameter und geben unterschiedliche Response-Felder zurück. Nicht alle oben im Playground aufgeführten Parameter funktionieren mit jedem Modell auf CometAPI.

Zustandsbehaftete Konversationen verwenden

Verketten Sie Responses mithilfe von previous_response_id, anstatt den Nachrichtenverlauf selbst zu verwalten:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cometapi.com/v1",
    api_key="<COMETAPI_KEY>",
)

# First turn
response = client.responses.create(
    model="gpt-5.4",
    input="What is quantum computing?",
)

# Second turn — previous context is included automatically
follow_up = client.responses.create(
    model="gpt-5.4",
    input="Can you explain that more simply?",
    previous_response_id=response.id,
)

print(follow_up.output_text)

Integrierte Tools verwenden

Die Responses API enthält von der Plattform bereitgestellte Tools, die keine Konfiguration erfordern:

Tool	Zweck
`web_search_preview`	Das Web nach Echtzeitinformationen durchsuchen
`file_search`	In hochgeladenen Dateien suchen
`code_interpreter`	Python-Code in einer Sandbox ausführen

Um ein integriertes Tool zu aktivieren, fügen Sie es zum Array tools hinzu:

response = client.responses.create(
    model="gpt-5.4",
    input="Find the current price of Bitcoin",
    tools=[{"type": "web_search_preview"}],
)

print(response.output_text)

Benutzerdefinierte Funktionen aufrufen

Definieren Sie Funktionen, die das Modell mit strukturierten Argumenten aufrufen kann:

response = client.responses.create(
    model="gpt-5.4",
    input="What's the weather in Tokyo?",
    tools=[{
        "type": "function",
        "name": "get_weather",
        "description": "Get current weather for a location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {"type": "string"}
            },
            "required": ["location"]
        }
    }],
)

Wenn das Modell eine Funktion aufruft, enthält das Array output in der Response ein Element vom Typ function_call mit dem Funktionsnamen und den geparsten Argumenten. Führen Sie die Funktion aus und senden Sie das Ergebnis in einer Follow-up-Request zurück.

Strukturierte Ausgabe anfordern

Um JSON-Ausgaben zu erzwingen, die einem bestimmten Schema entsprechen, verwenden Sie den Parameter text.format:

response = client.responses.create(
    model="gpt-5.4",
    input="List 3 programming languages with their main use cases",
    text={
        "format": {
            "type": "json_schema",
            "name": "languages",
            "strict": True,
            "schema": {
                "type": "object",
                "properties": {
                    "languages": {
                        "type": "array",
                        "items": {
                            "type": "object",
                            "properties": {
                                "name": {"type": "string"},
                                "use_case": {"type": "string"}
                            },
                            "required": ["name", "use_case"],
                            "additionalProperties": False
                        }
                    }
                },
                "required": ["languages"],
                "additionalProperties": False
            }
        }
    },
)

Configure reasoning

Für o-series- und GPT-5-Modelle steuern Sie die Tiefe des Reasoning mit reasoning.effort:

response = client.responses.create(
    model="o3",
    input="Solve this step by step: if f(x) = x^3 - 6x^2 + 11x - 6, find all roots.",
    reasoning={"effort": "high"},  # "low", "medium", or "high"
)

print(response.output_text)

Ein höherer Reasoning-Aufwand erzeugt gründlichere Antworten, verwendet aber mehr Tokens. Verwenden Sie "low" für einfache Anfragen und "high" für komplexe Probleme mit mehreren Schritten.

Stream responses

Um inkrementelle Ausgabe zu erhalten, setzen Sie stream auf true. Die API sendet Server-Sent Events (SSE) in dieser Reihenfolge:

response.created — Response-Objekt initialisiert
response.in_progress — Generierung gestartet
response.output_item.added — Neues Ausgabeelement (Nachricht oder Tool-Aufruf)
response.content_part.added — Inhaltsteil gestartet
response.output_text.delta — Textabschnitt (enthält das Feld delta)
response.output_text.done — Textgenerierung für diesen Inhaltsteil abgeschlossen
response.content_part.done — Inhaltsteil abgeschlossen
response.output_item.done — Ausgabeelement abgeschlossen
response.completed — Vollständige Response mit usage-Daten

Streamen Sie eine Response mit dem Python SDK:

stream = client.responses.create(
    model="gpt-5.4",
    input="Write a haiku about coding",
    stream=True,
)

for event in stream:
    if event.type == "response.output_text.delta":
        print(event.delta, end="")

Ausführliche Anleitungen zu den einzelnen Funktionen finden Sie in der OpenAI-Dokumentation: Text · Images · PDF files · Structured Outputs · Function Calling · Conversation State · Built-in Tools · Reasoning

Autorisierungen

Authorization

string

header

erforderlich

Bearer token authentication. Use your CometAPI key.

Body

application/json

model

string

erforderlich

Model ID to use for this request. See the Models page for current options.

Beispiel:

"gpt-5.4"

input

erforderlich

Text, image, or file inputs to the model, used to generate a response. Can be a simple string for text-only input, or an array of input items for multimodal content (images, files) and multi-turn conversations.

instructions

string

A system (or developer) message inserted into the model's context. When used with previous_response_id, instructions from the previous response are not carried over — this makes it easy to swap system messages between turns.

background

boolean

Standard:false

Whether to run the model response in the background. Background responses do not return output directly — you retrieve the result later via the response ID.

context_management

object[]

Context management configuration for this request. Controls how the model manages context when the conversation exceeds the context window.

Show child attributes

conversation

The conversation this response belongs to. Items from the conversation are prepended to input for context. Input and output items are automatically added to the conversation after the response completes. Cannot be used with previous_response_id.

include

enum<string>[]

Additional output data to include in the response. Use this to request extra information that is not included by default.

Verfügbare Optionen:

web_search_call.action.sources,

code_interpreter_call.outputs,

computer_call_output.output.image_url,

file_search_call.results,

message.input_image.image_url,

message.output_text.logprobs,

reasoning.encrypted_content

max_output_tokens

integer

An upper bound for the number of tokens that can be generated for a response, including visible output tokens and reasoning tokens.

max_tool_calls

integer

The maximum number of total calls to built-in tools that can be processed in a response. This limit applies across all built-in tool calls, not per individual tool. Any further tool call attempts by the model will be ignored.

metadata

object

Set of up to 16 key-value pairs that can be attached to the response. Useful for storing additional information in a structured format. Keys have a maximum length of 64 characters; values have a maximum length of 512 characters.

Show child attributes

parallel_tool_calls

boolean

Standard:true

Whether to allow the model to run tool calls in parallel.

previous_response_id

string

The unique ID of a previous response. Use this to create multi-turn conversations without manually managing conversation state. Cannot be used with conversation.

prompt

object

Reference to a prompt template and its variables.

Show child attributes

prompt_cache_key

string

A key used to cache responses for similar requests, helping optimize cache hit rates. Replaces the deprecated user field for caching purposes.

prompt_cache_retention

enum<string>

The retention policy for the prompt cache. Set to 24h to keep cached prefixes active for up to 24 hours.

Verfügbare Optionen:

in-memory,

24h

reasoning

object

Configuration options for reasoning models (o-series and gpt-5). Controls the depth of reasoning before generating a response.

Show child attributes

safety_identifier

string

A stable identifier for your end-users, used to help detect policy violations. Should be a hashed username or email — do not send identifying information directly.

Maximum string length: 64

service_tier

enum<string>

Specifies the processing tier for the request. When set, the response will include the actual service_tier used.

auto: Uses the tier configured in project settings (default behavior).
default: Standard pricing and performance.
flex: Flexible processing with potential cost savings.
priority: Priority processing with faster response times.

Verfügbare Optionen:

auto,

default,

flex,

priority

store

boolean

Standard:true

Whether to store the generated response for later retrieval via API.

stream

boolean

Standard:false

If set to true, the response data will be streamed to the client as it is generated using server-sent events (SSE). Events include response.created, response.output_text.delta, response.completed, and more.

stream_options

object

Options for streaming responses. Only set this when stream is true.

Show child attributes

temperature

number

Standard:1

Sampling temperature between 0 and 2. Higher values (e.g., 0.8) increase randomness; lower values (e.g., 0.2) make output more focused and deterministic. We recommend adjusting either this or top_p, but not both.

Erforderlicher Bereich: 0 <= x <= 2

text

object

Configuration for text output. Use this to request structured JSON output via JSON mode or JSON Schema.

Show child attributes

tool_choice

Standard:auto

Controls how the model selects which tool(s) to call.

auto (default): The model decides whether and which tools to call.
none: The model will not call any tools.
required: The model must call at least one tool.
An object specifying a particular tool to use.

tools

object[]

An array of tools the model may call while generating a response. CometAPI supports three categories:

Built-in tools: Platform-provided tools like web_search_preview and file_search.
Function calls: Custom functions you define, enabling the model to call your own code with structured arguments.
MCP tools: Integrations with third-party systems via MCP servers.

top_logprobs

integer

Number of most likely tokens to return at each position (0–20), each with an associated log probability. Must include message.output_text.logprobs in the include parameter to receive logprobs.

Erforderlicher Bereich: 0 <= x <= 20

top_p

number

Standard:1

Nucleus sampling parameter. The model considers tokens with top_p cumulative probability mass. For example, 0.1 means only the top 10% probability tokens are considered. We recommend adjusting either this or temperature, but not both.

Erforderlicher Bereich: 0 <= x <= 1

truncation

enum<string>

Standard:disabled

The truncation strategy for handling inputs that exceed the model's context window.

auto: The model truncates the input by dropping items from the beginning of the conversation to fit.
disabled (default): The request fails with a 400 error if the input exceeds the context window.

Verfügbare Optionen:

auto,

disabled

user

string

veraltet

Deprecated. Use safety_identifier and prompt_cache_key instead. A stable identifier for your end-user.

Antwort

200 - application/json

The generated Response object.

string

Unique identifier for the response.

Beispiel:

"resp_0a153ae8201f73bc0069a7e8044cc481"

object

enum<string>

The object type, always response.

Verfügbare Optionen:

response

Beispiel:

"response"

created_at

integer

Unix timestamp (in seconds) of when the response was created.

Beispiel:

1772611588

status

enum<string>

The status of the response.

Verfügbare Optionen:

completed,

in_progress,

failed,

cancelled,

queued

Beispiel:

"completed"

background

boolean

Whether the response was run in the background.

Beispiel:

false

completed_at

integer | null

Unix timestamp of when the response was completed, or null if still in progress.

Beispiel:

1772611589

error

object

Error information if the response failed, or null on success.

Show child attributes

incomplete_details

object

Details about why the response is incomplete, if applicable.

Show child attributes

instructions

string | null

The system instructions used for this response.

max_output_tokens

integer | null

The maximum output token limit that was applied.

model

string

The model used for the response.

Beispiel:

"gpt-4.1-nano"

output

object[]

An array of output items generated by the model. Each item can be a message, function call, or other output type.

Show child attributes

output_text

string

A convenience field containing the concatenated text output from all output message items.

parallel_tool_calls

boolean

Whether parallel tool calls were enabled.

previous_response_id

string | null

The ID of the previous response, if this is a multi-turn conversation.

reasoning

object

The reasoning configuration that was used.

Show child attributes

service_tier

string

The service tier actually used to process the request.

Beispiel:

"default"

store

boolean

Whether the response was stored.

temperature

number

The temperature value used.

Beispiel:

1

text

object

The text configuration used.

Show child attributes

tool_choice

The tool choice setting used.

tools

object[]

The tools that were available for this response.

top_p

number

The top_p value used.

Beispiel:

1

truncation

string

The truncation strategy used.

usage

object

Token usage statistics for this response.

Show child attributes

user

string | null

The user identifier, if provided.

metadata

object

The metadata attached to this response.

Show child attributes

content_filters

array | null

Content filter results applied to the response, if any.

frequency_penalty

number

Standard:0

The frequency penalty applied to the request.

max_tool_calls

integer | null

Maximum number of tool calls allowed, if set.

presence_penalty

number

Standard:0

The presence penalty applied to the request.

prompt_cache_key

string | null

Cache key for prompt caching, if applicable.

prompt_cache_retention

string | null

Prompt cache retention policy, if applicable.

safety_identifier

string | null

Safety system identifier for the response, if applicable.

top_logprobs

integer

Standard:0

Number of top log probabilities returned per token position.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cometapi.com/v1",
    api_key="<COMETAPI_KEY>",
)

response = client.responses.create(
    model="gpt-5.4",
    input="Tell me a three sentence bedtime story about a unicorn.",
)

print(response.output_text)

{
  "id": "resp_0a153ae8201f73bc0069a7e8044cc481",
  "object": "response",
  "created_at": 1772611588,
  "status": "completed",
  "background": false,
  "completed_at": 1772611589,
  "error": null,
  "incomplete_details": null,
  "instructions": null,
  "max_output_tokens": null,
  "model": "gpt-4.1-nano",
  "output": [
    {
      "id": "msg_0a153ae8201f73bc0069a7e8049a8881",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "type": "output_text",
          "annotations": [],
          "text": "Four."
        }
      ],
      "role": "assistant"
    }
  ],
  "parallel_tool_calls": true,
  "previous_response_id": null,
  "prompt_cache_key": null,
  "prompt_cache_retention": null,
  "reasoning": {
    "effort": null,
    "summary": null
  },
  "safety_identifier": null,
  "service_tier": "auto",
  "store": true,
  "temperature": 1,
  "text": {
    "format": {
      "type": "text"
    },
    "verbosity": "medium"
  },
  "tool_choice": "auto",
  "tools": [],
  "top_p": 1,
  "truncation": "disabled",
  "usage": {
    "input_tokens": 19,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 9,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 28
  },
  "user": null,
  "metadata": {}
}

Documentation Index

​Zustandsbehaftete Konversationen verwenden

​Integrierte Tools verwenden

​Benutzerdefinierte Funktionen aufrufen

​Strukturierte Ausgabe anfordern

​Configure reasoning

​Stream responses

Autorisierungen

Body

Antwort

Zustandsbehaftete Konversationen verwenden

Integrierte Tools verwenden

Benutzerdefinierte Funktionen aufrufen

Strukturierte Ausgabe anfordern

Configure reasoning

Stream responses