# EXAONE 4.5 33B (slim)

> EXAONE 4.5 33B is LG AI Research's open-weight vision-language model, 34B parameters, with tool calling, switchable reasoning and a 256K context, strong on documents and Korean. Its licence allows research and education only, with LG's own GGUF from 18 GB.

- Full: https://www.anchorterminal.com/tools/exaone-4-5-33b.md (~9,800 tokens) · this version ~1,830 tokens · JSON https://www.anchorterminal.com/tools/exaone-4-5-33b.json · canonical https://www.anchorterminal.com/tools/exaone-4-5-33b
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-10 · pricing https://www.anchorterminal.com/tools/exaone-4-5-33b#pricing · reviews https://www.anchorterminal.com/tools/exaone-4-5-33b#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 61/100 · rank #61 of 160 open-weight models · #51 in Open-weight models · not agent-ready · confidence low**


Assessment: A well-documented vision-language model with a full set of builds, tool calling, switchable reasoning, a 256K context and a stated knowledge cut-off. The licence decides whether you can use it: research and education only, output included, and LG can change the terms. Nothing answered on the repository and no safety results.

## Facts

- Kind: Open-weight model · vendor: LG AI Research · category: Open-weight models · legal entity: LG Management Development Institute Co., Ltd., the licensor named in the model's licence. The card names LG AI Research · provenance 49/100
- Packages: huggingface `LGAI-EXAONE/EXAONE-4.5-33B`, huggingface `LGAI-EXAONE/EXAONE-4.5-33B-GGUF`, huggingface `LGAI-EXAONE/EXAONE-4.5-33B-AWQ`, huggingface `LGAI-EXAONE/EXAONE-4.5-33B-FP8`
- Auth: None · pricing: Free · x402: no · licence: EXAONE AI Model License Agreement 1.2 - NC (research and education only)
- Benchmark probes: not run yet, so Performance is pending
- Surface graded: The open weights at huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B and the GGUF, AWQ and FP8 repositories beside it, read from the model card, the files and their metadata, the config, the chat template, the LICENSE, the commit history and the runtimes' own code at their latest releases on 10 November 2026. LG AI Research's site was read for who it is and gave little text. The technical report, the blog post and the GitHub repository weren't read. Nothing was downloaded or run
- Model: 34,350,097,664 parameters in the files. The card gives 31.7B for the language model and 1.29B for the vision encoder: 64 layers that interleave sliding-window and global attention, with one extra layer for multi-token prediction. LG's first open-weight vision-language model, built on the EXAONE 4.0 framework
- Files: Two safetensors shards in BF16 (68.7 GB), chat_template.jinja, config, processor and tokenizer files and the LICENSE. No pickle files and no Python. The GGUF repository has six precisions and three projector files. Hugging Face lists a SHA-256 for the large files
- Licence: EXAONE AI Model License Agreement 1.2 - NC, from LG Management Development Institute Co., Ltd. A revocable licence for research and educational use. Commercial use of the model, derivatives or output needs a separate agreement, and so does using it to develop models that compete with LG's. LG may amend the agreement at any time. Korean law governs
- Runtimes: The card documents TensorRT-LLM, vLLM, SGLang from a fork, and Transformers 5.8.0 or later, and says llama.cpp supports it. Released vLLM 0.33.0 and Transformers 5.22.2 have the exaone4_5 model, and llama.cpp b12242 has the architecture, with LG's own GGUF. SGLang 0.5.23 has no code for it
- Context: 262,144 tokens, per the card's configuration table and config.json
- Hardware: One H200, or four A100 40 GB cards with tensor parallelism, for the full 256K context, per the card. IQ4_XS with the projector is 20.6 GB
- Training data: Not described on the card beyond a knowledge cut-off of December 2024 and a line that LG tries to exclude personal, harmful and biased content. The technical report wasn't read
- Languages: English, Korean, Spanish, German, Japanese and Vietnamese, per the card's metadata
- Tool calling: The card's vLLM command sets auto tool choice with the hermes tool-call parser, and its agentic example drives the model through LangChain with MCP servers. Its results table has a row group for agentic tool use
- Reasoning: enable_thinking is on by default, and the card says to set it false for the non-reasoning mode. The qwen3 reasoning parser returns the reasoning apart from the answer
- Recommended settings: temperature 1.0, top_p 0.95 and presence_penalty 1.5 in general; temperature 0.6 with top_k 20 for OCR, documents and Korean; no presence penalty for text-only input. The card says the model prefers to answer in a boxed format

- Scores: Capability pending, Reach 65, Risk 50, Agent interface 85, Documentation and upkeep 50 · total over the 7 assessed categories
- Why:  · Reach, Memory 20, the smallest build at 4 bits is LG's IQ4_XS GGUF, 18.03 GB with the 2.57 GB projector, and 20.60 GB × 1.2 = 24.7 GB, within 32 GB… · Risk, File safety 25, safetensors and GGUF only, no .bin, .pt or .pth files and no Python in the repository. · Agent interface, Chat template 20, chat_template.jinja in the repository (5,930 characters). · Documentation and upkeep, Model card 20, Uses (the quickstart's question answering, document and agent examples), limitations, the architecture, the parameters, the c…
- Sources: 19, open questions: 6, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/exaone-4-5-33b.json
- Its vendor verifies the listing with `verify_listing` or `POST https://www.anchorterminal.com/api/v1/verify` (the full twin says how); no effect on the grade.
- Support: https://github.com/LG-AI-EXAONE/EXAONE-4.5; more in the full twin

## Before you call it

1. Don't use it, or anything it writes, commercially. The licence covers the output too
2. Serve it with vLLM and `--reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser hermes`
3. Set `enable_thinking` false for a direct answer. Thinking is on by default
4. Use temperature 1.0, top_p 0.95 and presence_penalty 1.5 in general, and temperature 0.6 with top_k 20 for documents and Korean
5. For llama.cpp, load a projector file from LG's GGUF repository with the weights for image input
6. Its knowledge stops at December 2024

## Connect

```bash
pip install "vllm>=0.33.0" "transformers>=5.8.0"
vllm serve LGAI-EXAONE/EXAONE-4.5-33B --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser hermes
```

```bash
# vLLM serves an OpenAI-compatible API at http://localhost:8000/v1. The card recommends temperature 1.0,
# top_p 0.95 and presence_penalty 1.5, and enable_thinking false for a direct answer
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 75.4 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 73.6 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 12B IT | BB | 72.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)