The story so far
Weights, every dated article, and repo commits on one rail.
8 repo commits since the weights landed
Teal is the weights (2026-04-04), grey ticks are repo commits.
EXAONE-4.5-33B
LGAI-EXAONE exaone out 7 months ago
34.4B parameters, images and text in, text out, 256K tokens of context.
Verified 2026-10-25 Rank #682 in LLMs · score 72.5 method v1 Sources 3 Downloads 30d 37.1k · #403 of 619 tracked
Can I use it?exaoneCustom license; read the repo LICENSE before commercial hostinglicence file2026-10-24
Do people use it?37.1kdownloads, 30 days#403 of 619 we track · below the median+1 likes, downloads −19% this week188 likesHugging Face2026-10-24
Where it's used
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 21 quantized builds and 2 Spaces.
- As a service0no API provider listed
- On own hardware21quantized builds
- Built on00 fine-tunes · 0 adapters · 0 merges
- In apps2Hugging Face Spaces
Top Spaces running it: mayafree/Model-Genome-Korea ♥ 33
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 64 GB Mac (unified memory) at INT4
About 21 GB at INT4, 41 GB at FP8, 82 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
What people built on it
21 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
21 derivatives on the Hub, most of them quantized
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
Benchmarks
16 rows
Benchmarks 16 rows
measured third-party leaderboard rowauthor-reported number from the model card or eval files (16)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported AI2D · Understand images and video | 89 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported BLINK · Understand images and video | 68.8 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported BabyVision | 18.8 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported CharXiv · Extract text from documents | 71.7 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported ChartQA · Understand images and video | 62.2 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported HallusionBench · Understand images and video | 63.7 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported LogicVista | 73.8 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MATH-500 · Reason through hard problems | 79.1 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MMMU · Understand images and video | 78.7 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MMMU-Pro · Understand images and video | 68.6 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MMStar · Understand images and video | 74.9 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MathVision · Understand images and video | 75.2 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MathVista · Understand images and video | 85 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported MedXpertQA-MM | 42.1 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported OCRBench · Extract text from documents | 63.2 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
| author-reported OmniDocBench · Extract text from documents | 81.2 | EXAONE 4.5 33B (Reasoning) | model author | 2026-10-07 |
Use-case evidence
6 use cases
What it is good at 6 use cases
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
- Understand images and videoauthor-reportedAuthor reports MMMU 78.7Author reports MMMU-Pro 68.6Author reports MathVision 75.2+10 more
- Extract text from documentsauthor-reportedAuthor reports CharXiv (RQ) 71.7Author reports OCRBench v2 63.2Author reports OmniDocBench v1.5 81.2
- Reason through hard problemsauthor-reportedAuthor reports WeMath 79.1chat template: <think>
- Chat and follow instructionsstructuraldecision-model estimate p=0.82tag: conversationalchat template present
- Read long documentsstructuraldecision-model estimate p=1.00config: 262,144 context
- Call tools and APIsstructuralchat template: <tool_call>
Facts
22 fields
Listed in LLMs
Facts
- EXAONE-4.5-33B has 34.4B parameters. (safetensors metadata, observed 2026-10-24)
- Context length is 262,144 tokens (256K). (config.json max_position_embeddings, observed 2026-10-24)
- License: exaone (friction: review). Custom license; read the repo LICENSE before commercial hosting. (license file, observed 2026-10-24)
- Created 2026-04-04, last modified 2026-07-06. (HF API, observed 2026-10-24)
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Exaone4_5_ForConditionalGeneration |
| Model type | exaone4_5 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 34,350,097,664 |
| Size band | L |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| Languages | en, ko, es, de, ja, vi |
| Papers | arXiv:2604.08644 |
| License | exaone (other) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · Exaone4_5_ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Hardware fit
3 estimates
Hardware fit ESTIMATE
Weights-only memory for EXAONE-4.5-33B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 82 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 41 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 21 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
21 variants
Lineage
quantized: 21
Top variants (21 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| LGAI-EXAONE/EXAONE-4.5-33B-AWQ | quantized | AWQ | 7.51B | M | other | LGAI-EXAONE | 7 | 143.9k |
| LGAI-EXAONE/EXAONE-4.5-33B-FP8 | quantized | -FP8 | 34.4B | L | other | LGAI-EXAONE | 27 | 93.8k |
| LGAI-EXAONE/EXAONE-4.5-33B-GGUF | quantized | GGUF | — | — | other | LGAI-EXAONE | 54 | 28.5k |
| mradermacher/EXAONE-4.5-33B-i1-GGUF | quantized | GGUF | — | — | other | mradermacher | 2 | 4.8k |
| mradermacher/EXAONE-4.5-33B-GGUF | quantized | GGUF | — | — | other | mradermacher | 0 | 2.2k |
| jkim96/EXAONE-4.5-33B-DASHQ-Q2-GGUF | quantized | GGUF | — | — | other | jkim96 | 0 | 1.1k |
| jkim96/EXAONE-4.5-33B-DASHQ-INT3-g128 | quantized | — | 2.86B | M | other | jkim96 | 0 | 942 |
| jkim96/EXAONE-4.5-33B-DASHQ-INT3-g64 | quantized | — | 2.86B | M | other | jkim96 | 0 | 668 |
| jkim96/EXAONE-4.5-33B-DASHQ-INT2-g32 | quantized | — | 2.86B | M | other | jkim96 | 0 | 637 |
| jkim96/EXAONE-4.5-33B-DASHQ-INT3-g32 | quantized | — | 2.86B | M | other | jkim96 | 0 | 613 |
| sashisuseso/EXAONE-4.5-33B-GGUF | quantized | GGUF | — | — | other | sashisuseso | 1 | 556 |
| lee5j/EXAONE-4.5-33B-NVFP4 | quantized | NVFP4 | 20.2B | L | other | lee5j | 1 | 470 |
| jkim96/EXAONE-4.5-33B-DASHQ-INT4-g64 | quantized | INT4 | 2.86B | M | other | jkim96 | 0 | 222 |
| haanjack/EXAONE-4.5-33B-MXFP4 | quantized | MXFP4 | 20.9B | L | other | haanjack | 0 | 206 |
| jkim96/EXAONE-4.5-33B-DASHQ-INT4-g32 | quantized | INT4 | 2.86B | M | other | jkim96 | 0 | 191 |
| jkim96/EXAONE-4.5-33B-DASHQ-INT4-g128 | quantized | INT4 | 2.86B | M | other | jkim96 | 0 | 183 |
| Hyun9junn/EXAONE-4.5-33B-FlatQuant-W4A16 | quantized | -W4A16 | 6.98B | M | other | Hyun9junn | 0 | 182 |
| lee5j/EXAONE-4.5-33B_GPTQ4 | quantized | GPTQ | 34.6B | L | — | lee5j | 0 | 126 |
| lee5j/EXAONE-4.5-33B_GPTQ8 | quantized | GPTQ | 34.4B | L | — | lee5j | 0 | 108 |
| GotoAI-Inc/EXAONE-4.5-33B-W8A16 | quantized | — | 34.4B | L | other | GotoAI-Inc | 0 | 55 |
| GotoAI-Inc/EXAONE-4.5-33B-W4A16 | quantized | -W4A16 | 34.4B | L | other | GotoAI-Inc | 0 | 30 |
Repository changelog
8 commits
Repository changelog 8 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-07-06 | Update dependencies to official releases | 570aa4b |
| 2026-06-16 | Update config.json | e9949a4 |
| 2026-04-29 | Update config.json | c8645dd |
| 2026-04-29 | Update README.md | 04a56cc |
| 2026-04-15 | Update README.md | 58d6616 |
| 2026-04-14 | Update README.md | 49c2d11 |
| 2026-04-09 | Update README.md | 9644e54 |
| 2026-04-09 | Initial commit | acb250c |
Papers
1 paper
Papers 1
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
- EXAONE 4.5 Technical Report this model
Similar models
3 models
Similar models
- gemma-4-31B-it · 31.3B same size band (L), same task (image-text-to-text), more providers, permissive license
- Qwen3.8-27B · 27.8B same size band (L), same task (image-text-to-text), more providers, permissive license
- Qwen2.5-VL-32B-Instruct · 33.5B same size band (L), same task (image-text-to-text), more providers, permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:42 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:13 UTC · Last verified 2026-10-25
Follow EXAONE-4.5-33B
Watch this model
- /feeds/models/LGAI-EXAONE--EXAONE-4.5-33B.xml(RSS 2.0) · every change to EXAONE-4.5-33B
- /feeds/families/LGAI-EXAONE--EXAONE-4.5.xml(RSS 2.0) · every change across the EXAONE-4.5 family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.