K-EXAONE 2.5

POST /v1/chat/completions

The K-EXAONE 2.5 model is available through the EXAONE API Platform. The EXAONE API is OpenAI-compatible.

K-EXAONE 2.5 is a text-only model.

The maximum context length is 524,288 tokens. If the number of input tokens plus max_tokens exceeds this limit, the API returns a 400 error.

Request

Field	Type	Description	Required
model	string	`k-exaone_v2.5`	mandatory
messages	array<object>	List of messages in the conversation	mandatory
messages[].role	string	system / user / assistant	mandatory
messages[].content	string	Content of the message: a prompt, query, or response	mandatory
max_tokens	int	Maximum number of tokens to generate, including the reasoning process. Input tokens + max_tokens ≤ 524,288	optional
stream	boolean	Whether to stream the response (default: `false`)	optional
stream_options.include_usage	boolean	Whether to include a token usage chunk at the end of the stream. Available only when `stream` is `true`. (Default: `false`)	optional
temperature	float	Controls the randomness of the output by adjusting how sharp the probability distribution is. Higher values produce more random output. Range: 0.0 or higher. (Default: 1.0)	optional
top_p	float	Samples only from the smallest set of top tokens whose cumulative probability is within p. (Range: 0.0~1.0, Default: 0.95)	optional
top_k	int	Samples only from the k most likely tokens. Range: -1, or 1 or higher. Set to -1 to disable filtering. (Default: -1)	optional
presence_penalty	float	Penalizes tokens that have already appeared, which reduces repetition and encourages new topics. Positive values make repeated tokens less likely. (Range: -2.0~2.0. Default: 0.0)	optional
reasoning_effort	string	Reasoning (thinking) intensity: `low` / `high` / `max`. Higher values produce longer reasoning (default: `max`)	optional
chat_template_kwargs.preserve_thinking	boolean	Whether to keep the reasoning content of previous assistant turns in the input for multi-turn conversations. If `false`, only the reasoning after the last user message is kept (default: `false`)	optional

Recommended parameter settings

For the best model performance, we recommend the following settings.- General use: temperature=1.0, top_p=0.95- Agent use, or multi-turn conversations that keep previous reasoning: temperature=1.0, top_p=0.95, chat_template_kwargs.preserve_thinking=true

Reasoning intensity(reasoning_effort)

K-EXAONE 2.5 always reasons before answering, so enable_thinking is automatically set to true and can’t be disabled. Use reasoning_effort to adjust the reasoning intensity instead. We recommend low when a fast response is needed, high for difficult problems such as math and coding, and max for highly complex problems where accuracy matters more than response time.The reasoning process is returned separately in message.reasoning_content, and the final answer in message.content. The number of tokens used for reasoning is reported in usage.reasoning_tokens.

Leading newline in the output

content may start with a newline ("\n\n"). If you need to compare the string exactly, strip leading and trailing whitespace before use.

Response(stream: false)

Field	Type	Description
body	object	A chat completion object in the standard OpenAI format

Response(stream: true)

Field	Type	Description
data	string	An OpenAI chat completion chunk sent over SSE

Example request(stream: false)

curl 'https://api.exaone.ai/v1/chat/completions' \
  -H "Authorization: Bearer $EXAONE_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "k-exaone_v2.5",
  "max_tokens": 512,
  "messages": [
    {
      "role": "user",
      "content": "Hi, EXAONE"
    }
  ],
  "stream": false
}'

import requests

url = "https://api.exaone.ai/v1/chat/completions"
headers = {
    "Authorization": f"Bearer {EXAONE_API_KEY}",
    "Content-Type": "application/json",
}
payload = {
    "model": "k-exaone_v2.5",
    "messages": [{"role": "user", "content": "Hi, EXAONE"}],
    "stream": False,
}

response = requests.post(url=url, headers=headers, json=payload)
print(response.text)

(async () => {
  const url = 'https://api.exaone.ai/v1/chat/completions';
  const headers = {
    Authorization: `Bearer ${process.env.EXAONE_API_KEY}`,
    'Content-Type': 'application/json',
  };
  const payload = {
    model: 'k-exaone_v2.5',
    max_tokens: 512,
    messages: [{ role: 'user', content: 'Hi, EXAONE' }],
    stream: false,
  };

  try {
    const response = await fetch(url, {
      method: 'POST',
      headers: headers,
      body: JSON.stringify(payload),
    });
    console.log(await response.text());
  } catch (error) {
    console.error('Error:', error);
  }
})();

Example response(stream: false)

{
  "id": "1f30a3635ebd4e4cb57e4d41e2f18d55",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null,
      "message": {
        "content": "\n\nHello! How can I help you today?",
        "refusal": null,
        "role": "assistant",
        "annotations": null,
        "audio": null,
        "function_call": null,
        "tool_calls": null,
        "reasoning_content": "We need answer to user \"Hi, EXAONE\". We are AI assistant accessed via API. Need respond. We shouldn't claim EXAONE? We are likely not EXAONE, unless system? The system prompt is how am I supposed to behave, no identity. User says Hi, EXAONE. We can greet back, maybe clarify: I'm EXAONE? Need be honest. We are not necessarily called EXAONE. However no explicit model identity. In API, maybe they are addressing assistant. We can say \"Hi! I'm an AI assistant. How can I help?\" If this is a test? They said \"Hi, EXAONE\", maybe expect assistant to identify as EXAONE? We need not misrepresent. If team? There is no identity, so safest: \"Hi! I don't have a name, but happy to help.\" But user might be talking to EXAONE model (LG AI Research). We shouldn't claim if not. Maybe \"Hello! How can I assist you today?\" Simple. Could mention \"I can try.\" Since asked just hi, no need. Final answer.\n"
      },
      "matched_stop": 53
    }
  ],
  "created": 1793945790,
  "model": "k-exaone_v2.5",
  "object": "chat.completion",
  "metadata": {
    "weight_version": "default",
    "weight_versions": [{ "version": "default", "start": 0, "end": 218 }]
  },
  "moderation": null,
  "service_tier": null,
  "system_fingerprint": null,
  "usage": {
    "completion_tokens": 218,
    "prompt_tokens": 22,
    "total_tokens": 240,
    "completion_tokens_details": null,
    "prompt_tokens_details": null,
    "reasoning_tokens": 209
  }
}

Example request(stream: true)

curl 'https://api.exaone.ai/v1/chat/completions' \
  -H "Authorization: Bearer $EXAONE_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "k-exaone_v2.5",
  "max_tokens": 512,
  "messages": [
    {
      "role": "user",
      "content": "Hi, EXAONE"
    }
  ],
  "stream": true
}'

import requests

url = "https://api.exaone.ai/v1/chat/completions"
headers = {
    "Authorization": f"Bearer {EXAONE_API_KEY}",
    "Content-Type": "application/json",
}
payload = {
    "model": "k-exaone_v2.5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "Hi, EXAONE"}],
    "stream": True,
}

with requests.post(url=url, headers=headers, json=payload, stream=True) as response:
    for line in response.iter_lines(decode_unicode=True):
        if line:
            print(line)

(async () => {
  const url = 'https://api.exaone.ai/v1/chat/completions';
  const headers = {
    Authorization: `Bearer ${process.env.EXAONE_API_KEY}`,
    'Content-Type': 'application/json',
  };
  const payload = {
    model: 'k-exaone_v2.5',
    max_tokens: 512,
    messages: [{ role: 'user', content: 'Hi, EXAONE' }],
    stream: true,
  };

  try {
    const response = await fetch(url, {
      method: 'POST',
      headers: headers,
      body: JSON.stringify(payload),
    });

    const decoder = new TextDecoder();
    for await (const chunk of response.body) {
      process.stdout.write(decoder.decode(chunk, { stream: true }));
    }
  } catch (error) {
    console.error('Error:', error);
  }
})();

Example response(stream: true)

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"content": "", "role": "assistant"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"reasoning_content": "We"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"reasoning_content": " need"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"reasoning_content": " answer to"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"reasoning_content": " user. They said \"Hi"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

...

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"content": "how"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {"content": " can I help you today? 😊"}, "index": 0}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [{"delta": {}, "finish_reason": "stop", "index": 0, "matched_stop": 53}], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk"}

data: {"id": "2fa2bf0467cf4d758fe4c4dba28f2b18", "choices": [], "created": 1793947427, "model": "k-exaone_v2.5", "object": "chat.completion.chunk", "usage": {"completion_tokens": 234, "prompt_tokens": 22, "total_tokens": 256, "reasoning_tokens": 216}}

data: [DONE]