South Korea33B parametersImage + videoReasoning + tools

Explore EXAONE 4.5 33B

Use LG AI Research's popular multimodal model for Korean and English visual reasoning, document analysis, OCR, text generation, and agentic workflows.

Model

LGAI-EXAONE/EXAONE-4.5-33B

Input

Configure your request

messages

The OpenAI-compatible chat messages array. The model’s processor accepts text, image, and video content, plus assistant tool calls and tool-result messages; the documented vLLM configuration permits up to 64 images per prompt.

tools Optional

Optional OpenAI-compatible JSON array of function tools accepted by the model’s chat template.

tool_choice Optional

OpenAI-compatible tool-selection mode.

Parameters

Generation settings

Maximum output tokens

Temperature

LG’s recommended general-purpose and text-only setting.

Top P

LG’s recommended setting.

Top K

Recommended for OCR, documents, and Korean input.

Presence penalty

LG’s recommended multimodal setting.

Enable thinkingPassed as chat_template_kwargs.enable_thinking; enabled by default for EXAONE 4.5.

Explore EXAONE 4.5 33B

Configure your request

Response