# K-EXAONE 2.5 - Core.Today AI API

> LG AI Research's K-EXAONE 2.5 over the direct EXAONE API - a Korean/English reasoning model (750B total, 31B active parameters) with a 512K-token context window, tool calling and structured outputs. Currently available for non-commercial use only, for performance evaluation. It always reasons before answering, so keep max_tokens at 2,048 or more.

- **Provider**: LG AI Research
- **Model ID**: k-exaone_v2.5
- **Category**: LLM
- **Credits**: 0.56 per 1K tokens (avg)
- **Speed**: Medium
- **Quality**: High

## Model Specifications
- **Context Window**: 524K tokens
- **Max Output**: 524K tokens
- **Training Cutoff**: 2026-01
- **Supported Formats**: text, json, markdown
- **Compatible SDK**: OpenAI

### Capabilities
- Function Calling
- Streaming
- JSON Mode
- System Prompt

### Token Pricing (per 1M tokens)
- **Input Tokens**: 185.8 credits ($0.12)
- **Output Tokens**: 929 credits ($0.62)

## Features
- 512K-token context window
- Korean and English reasoning model
- Tool calling and structured outputs
- reasoning_effort low / high / max
- OpenAI-compatible API (chat completions) via /llm/exaone/v1

## Use Cases
- Korean-language reasoning and long-document analysis
- Agent workflows with tool calling
- Structured extraction from Korean text

## API Endpoint

Base URL: https://api.core.today
Endpoint: POST /llm/exaone/v1/chat/completions

## Authentication
Header: Authorization: Bearer YOUR_API_KEY

Note: LLM endpoints use OpenAI-compatible format with Authorization Bearer token.

## Input Parameters

### Required
- **messages**: array - Chat messages in OpenAI format (system/user/assistant roles).

### Optional
- **max_tokens**: integer (default: 2048) - Maximum completion tokens including reasoning. Keep >= 2048 or the answer may be cut off before any content is produced.
- **reasoning_effort**: string (default: max) - Reasoning depth: low, high or max.
- **temperature**: number (default: 1.0) - Sampling temperature. LG recommends 1.0 with top_p 0.95.
- **stream**: boolean (default: false) - Stream the response as server-sent events.

## Examples

### Chat completion
OpenAI-compatible chat completion over the direct EXAONE route
```json
{
  "model": "k-exaone_v2.5",
  "messages": [
    {
      "role": "user",
      "content": "다음 보고서를 다섯 줄로 요약해 줘."
    }
  ],
  "max_tokens": 4096
}
```

## Response Format
```json
{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Response text here"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 100,
    "completion_tokens": 50,
    "total_tokens": 150
  }
}
```

## Tips
- Non-commercial use only (performance evaluation) - charges are an API infrastructure usage fee, not a for-profit price
- Works with the OpenAI SDK - set base_url to https://api.core.today/llm/exaone/v1
- Keep max_tokens >= 2048: reasoning tokens come first and are billed at the output rate
- The answer may start with a blank line (\n\n) - trim it before exact comparisons
- Reasoning text is returned in reasoning_content, the answer in content

## Documentation
https://docs.exaone.ai/ko/docs/k-exaone-2.5