Responses API

POST https://api.shannon-ai.com/v1/responses

The smallest request is a model id and a text. The reply is not streamed unless you ask for it.

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.shannon-ai.com/v1",
)

response = client.responses.create(
    model="shannon-3",
    input="Say hello in one sentence.",
)

print(response.output_text)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://api.shannon-ai.com/v1",
});

const response = await client.responses.create({
  model: "shannon-3",
  input: "Say hello in one sentence.",
});

console.log(response.output_text);

curl https://api.shannon-ai.com/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "shannon-3",
    "input": "Say hello in one sentence."
  }'

The reply is a response object. Its output holds the reasoning, when the model reasoned, and the message:

{
  "id": "resp_5f0c1e7a9b3d4c62a8e1f07d2b46c9a3",
  "object": "response",
  "created_at": 1791625200,
  "model": "shannon-3",
  "status": "completed",
  "error": null,
  "incomplete_details": null,
  "instructions": null,
  "metadata": {},
  "parallel_tool_calls": true,
  "temperature": null,
  "tool_choice": "auto",
  "tools": [],
  "top_p": null,
  "output": [
    {
      "id": "rs_8a2e6c1f4b7d4e03b9c5a1d70f3e2b64",
      "type": "reasoning",
      "summary": [
        {
          "type": "summary_text",
          "text": "The user wants a greeting in one sentence. Keep it short and friendly."
        }
      ]
    },
    {
      "id": "msg_1c7b9e3a5d2f4a86b0e4c8d19a6f7e25",
      "type": "message",
      "role": "assistant",
      "status": "completed",
      "content": [
        {
          "type": "output_text",
          "text": "Hello, it is good to meet you.",
          "annotations": []
        }
      ]
    }
  ],
  "usage": {
    "input_tokens": 1184,
    "output_tokens": 46,
    "total_tokens": 1230
  }
}

Request fields

model and input are what a request needs. The Applied by column names the models on which a field changes the reply. The hosted open-weight models are the twelve ids of the model list; the Shannon 3 family is shannon-3, shannon-3-pro, shannon-3.1 and shannon-3.1-pro. Models & pricing

previous_response_id, store, metadata, parallel_tool_calls, truncation, include, user and prompt_cache_key are accepted so that existing client code runs unchanged. They do not change the reply. Because no conversation is stored, a follow-up request carries the earlier turns in input.

Input items

An input item is either a message, with a role and a content, or a typed item that carries a tool call or its result. content is a string or an array of parts.

Put instructions in the instructions field. With a Shannon 3 family id, put instructions that must hold into the user message.

The items of a reply's output can go back into input as they are: message and function_call items are read, reasoning items are skipped.

Content parts

On the Shannon tiers images and files are read from the last user item of input. Sizes and limits have their own page. Images and files

A request with instructions and options

This request sends instructions, a conversation as input items, an output limit, a temperature and a reasoning effort. It uses a hosted open-weight model, which applies all of them.

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.shannon-ai.com/v1",
)

response = client.responses.create(
    model="DeepSeek-V4-Flash-0731-W4A16-AUTOROUND-REAP",
    instructions="You are a physics teacher. Answer in two sentences.",
    input=[
        {"role": "user", "content": "Why is the sky blue?"},
        {"role": "assistant", "content": "Air scatters blue light more than red light."},
        {"role": "user", "content": "And why are sunsets red?"},
    ],
    max_output_tokens=512,
    temperature=0.3,
    reasoning={"effort": "low"},
)

for item in response.output:
    if item.type == "reasoning":
        print("reasoning:", item.summary[0].text)
    elif item.type == "message":
        print("answer:", item.content[0].text)
print(response.usage)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://api.shannon-ai.com/v1",
});

const response = await client.responses.create({
  model: "DeepSeek-V4-Flash-0731-W4A16-AUTOROUND-REAP",
  instructions: "You are a physics teacher. Answer in two sentences.",
  input: [
    { role: "user", content: "Why is the sky blue?" },
    { role: "assistant", content: "Air scatters blue light more than red light." },
    { role: "user", content: "And why are sunsets red?" },
  ],
  max_output_tokens: 512,
  temperature: 0.3,
  reasoning: { effort: "low" },
});

for (const item of response.output) {
  if (item.type === "reasoning") console.log("reasoning:", item.summary[0].text);
  if (item.type === "message") console.log("answer:", item.content[0].text);
}
console.log(response.usage);

curl https://api.shannon-ai.com/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "DeepSeek-V4-Flash-0731-W4A16-AUTOROUND-REAP",
    "instructions": "You are a physics teacher. Answer in two sentences.",
    "input": [
      {"role": "user", "content": "Why is the sky blue?"},
      {"role": "assistant", "content": "Air scatters blue light more than red light."},
      {"role": "user", "content": "And why are sunsets red?"}
    ],
    "max_output_tokens": 512,
    "temperature": 0.3,
    "reasoning": {"effort": "low"}
  }'

The reply of a hosted open-weight model has id, object, model, created_at, status, output and usage. Its usage adds the tokens read from cache and the tokens spent on reasoning:

{
  "usage": {
    "input_tokens": 58,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 67,
    "output_tokens_details": {
      "reasoning_tokens": 19
    },
    "total_tokens": 125
  }
}

The response object

id, object, status, output and usage are on every reply. Build your code on those.

Output items

The OpenAI SDKs join the text of the message for you in response.output_text.

Streaming

Set stream to true for server-sent events. Each frame has an event: line with the name of the event and a data: line with a JSON object; the object repeats the name in type. Match deltas to their item by output_index.

With the OpenAI SDK, read the events by their type:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.shannon-ai.com/v1",
)

stream = client.responses.create(
    model="shannon-3",
    input="Explain server-sent events in three sentences.",
    stream=True,
)

for event in stream:
    kind = getattr(event, "type", "error")     # an event without a type is an error event
    if kind == "response.reasoning_summary_text.delta":
        pass                                   # a piece of reasoning: event.delta
    elif kind == "response.output_text.delta":
        print(event.delta, end="", flush=True)
    elif kind == "response.output_item.done" and event.item.type == "function_call":
        print("\ntool call:", event.item.name, event.item.arguments)
    elif kind == "response.completed":
        print("\n", event.response.usage)
    elif kind in ("response.failed", "error"):
        raise RuntimeError("the stream failed, send the request again")

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://api.shannon-ai.com/v1",
});

const stream = await client.responses.create({
  model: "shannon-3",
  input: "Explain server-sent events in three sentences.",
  stream: true,
});

for await (const event of stream) {
  if (event.type === "response.reasoning_summary_text.delta") {
    // a piece of reasoning: event.delta
  } else if (event.type === "response.output_text.delta") {
    process.stdout.write(event.delta);
  } else if (event.type === "response.output_item.done" && event.item.type === "function_call") {
    console.log("\ntool call:", event.item.name, event.item.arguments);
  } else if (event.type === "response.completed") {
    console.log("\n", event.response.usage);
  } else if (event.type === "response.failed" || event.type === "error") {
    throw new Error("the stream failed, send the request again");
  }
}

curl -N https://api.shannon-ai.com/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "shannon-3",
    "input": "Explain server-sent events in three sentences.",
    "stream": true
  }'

The events arrive in this order. The groups for reasoning, the message and tool calls appear only when the reply has them.

event: response.created
data: {"type":"response.created","response":{"id":"resp_5f0c…","object":"response","status":"in_progress","output":[], …}}

event: response.in_progress
data: {"type":"response.in_progress","response":{"id":"resp_5f0c…","object":"response","status":"in_progress","output":[], …}}

event: response.output_item.added
data: {"type":"response.output_item.added","output_index":0,"item":{"id":"msg_1c7b…","type":"message","role":"assistant","status":"in_progress","content":[]}}

event: response.content_part.added
data: {"type":"response.content_part.added","output_index":0,"content_index":0,"part":{"type":"output_text","text":""}}

event: response.output_text.delta
data: {"type":"response.output_text.delta","output_index":0,"content_index":0,"delta":"Server-sent"}

event: response.output_text.delta
data: {"type":"response.output_text.delta","output_index":0,"content_index":0,"delta":" events let a server push text over one HTTP reply."}

event: response.output_text.done
data: {"type":"response.output_text.done","output_index":0,"content_index":0,"text":"Server-sent events let a server push text over one HTTP reply."}

event: response.content_part.done
data: {"type":"response.content_part.done","output_index":0,"content_index":0,"part":{"type":"output_text","text":"Server-sent events let a server push text over one HTTP reply."}}

event: response.output_item.done
data: {"type":"response.output_item.done","output_index":0,"item":{"id":"msg_1c7b…","type":"message","role":"assistant","status":"completed","content":[{"type":"output_text","text":"Server-sent events let a server push text over one HTTP reply.","annotations":[]}]}}

event: response.completed
data: {"type":"response.completed","response":{"id":"resp_5f0c…","object":"response","status":"completed","output":[…],"usage":{"input_tokens":1190,"output_tokens":18,"total_tokens":1208}, …}}

A stream that fails

On the Shannon tiers a failed reply ends with response.failed in place of response.completed. Its response.status is failed and response.error holds a code, rate_limit_exceeded or server_error, and a message.

On the hosted open-weight models a failure is an error event with code api_error and a message. response.completed still follows it.

Treat a stream as failed when it holds a response.failed or an error event, and send the request again.

Tools

A tool is a function your code runs. Describe it in tools; when the model wants it, the reply holds a function_call item in place of, or next to, the message.

A tool definition in tools:

[
  {
    "type": "function",
    "name": "get_weather",
    "description": "Current weather for a city.",
    "parameters": {
      "type": "object",
      "properties": {
        "city": {
          "type": "string"
        }
      },
      "required": [
        "city"
      ]
    }
  }
]

The output of a reply that calls it:

[
  {
    "id": "call_3d9a7c1e5b2f4a60c8e1d7f09b24a6c5",
    "type": "function_call",
    "name": "get_weather",
    "call_id": "call_3d9a7c1e5b2f4a60c8e1d7f09b24a6c5",
    "arguments": "{\"city\":\"Paris\"}"
  }
]

Run the function, then send the request again with the call and its result added to input:

[
  {
    "role": "user",
    "content": "What is the weather in Paris?"
  },
  {
    "type": "function_call",
    "call_id": "call_3d9a7c1e5b2f4a60c8e1d7f09b24a6c5",
    "name": "get_weather",
    "arguments": "{\"city\":\"Paris\"}"
  },
  {
    "type": "function_call_output",
    "call_id": "call_3d9a7c1e5b2f4a60c8e1d7f09b24a6c5",
    "output": "{\"temperature_c\": 18, \"sky\": \"cloudy\"}"
  }
]

Only tools of type function are read. On shannon-1.6-*, shannon-2-* and shannon-coder-1 the parameters schema is reduced to the common JSON Schema keywords: oneOf is read as anyOf, and keywords such as $ref, additionalProperties, const, default and title are not passed to the model.

tool_choice values required and a named function are applied by the hosted open-weight models.

The tool loop is described in full on its own page. Function Calling

Structured output

This format spells structured output as text.format. The schema sits directly in format, next to name and strict, not under a json_schema member as on Chat Completions.

• {"type": "json_object"} asks for any JSON object.
• {"type": "json_schema", "name", "schema", "strict"} asks for JSON that follows your schema. strict is true when you leave it out.

{
  "model": "shannon-1.6-lite",
  "input": "Extract the city and the date: 'The conference opens in Lisbon on 12 May 2027.'",
  "text": {
    "format": {
      "type": "json_schema",
      "name": "event",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "city": {
            "type": "string"
          },
          "date": {
            "type": "string",
            "description": "ISO 8601 date"
          }
        },
        "required": [
          "city",
          "date"
        ]
      }
    }
  }
}

The JSON arrives as text in the output_text part of the message; parse it yourself. A hosted open-weight model without structured output returns 400.

{"city": "Lisbon", "date": "2027-05-12"}

Which model takes which form is listed on its own page. Structured Outputs

Reasoning

A model that reasons returns its reasoning as an output item of type reasoning, before the message. In a stream it arrives as response.reasoning_summary_text.delta events.

The hosted open-weight models apply reasoning.effort: off, low, medium or high, with high when you send nothing. off returns no reasoning item. Send it as an object; a bare string is not read on this endpoint.

The efforts are described on their own page. Reasoning effort

Output limit

max_output_tokens is 4,096 when you do not send it. It is the number of tokens set aside from your balance when the request starts; when the reply is complete, that amount is replaced by what the request used. A value that does not fit into your remaining balance returns 429 Quota exceeded.

It limits the length of the reply on shannon-1.6-*, shannon-2-* and shannon-coder-1, and the length of the answer text on the hosted open-weight models.

On this endpoint shannon-coder-1 is charged in tokens from your balance, like every other model.

A request body can be up to 32 MiB. Limits and balance

Errors

Errors use the OpenAI error body: an error object with type and message; some also carry code and param with the value null. The table lists what this endpoint returns most often. The full list, with what to retry, has its own page. Error Handling