EXAONE-3.5-2.4B-Instruct

2.41B parameters, text in, text out, 32K tokens of context.

Verified 2026-10-25 Rank #690 in LLMs · score 72.3 method v1 Sources 3 Downloads 30d 54k · #378 of 619 tracked

Can I run it?1.4GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?exaoneCustom license; read the repo LICENSE before commercial hostinglicence file2026-10-24

Do people use it?54kdownloads, 30 days#378 of 619 we track · below the median−1 likes, downloads +6% this week197 likesHugging Face2026-10-24

Worth reading

Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.

Latest · 23 months ago

Open-sourcing Three EXAONE 3.5 Models: 2.4B, 7.8B, 32B

LG AI Research released three EXAONE 3.5 models including a 2.4B, 7.8B, and 32B variant.

lgresearch.ai 2024-12-09 13 HN pts · 4 comments 1.2k words about the EXAONE 3.5 family family

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: 23 quantized builds and 23 Spaces.

As a service0no API provider listed
On own hardware23quantized builds
Built on9767 fine-tunes · 30 adapters · 0 merges
In apps23Hugging Face Spaces

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 1.4 GB at INT4, 2.9 GB at FP8, 5.8 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

1 dated article and 4 repo commits so far

Teal is the weights (2024-12-01), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.

What people built on it

120 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

120 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF lmstudio-community/EXAONE-3.5-2.4B-Instruct-GGUF for llama.cpp, Ollama, LM Studio 83.3k downloads MLX alexgusevski/EXAONE-3.5-2.4B-Instruct-q8-mlx for Apple silicon 230 downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

7 rows

Benchmarks 7 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (7)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 AlpacaEval · Chat and follow instructions	37.1	EXAONE 3.5 2.4B	model author	2026-10-07
author-reported Arena-Hard · Chat and follow instructions	48.2	EXAONE 3.5 2.4B	model author	2026-10-07
author-reported IFEval · Chat and follow instructions	73.6	EXAONE 3.5 2.4B	model author	2026-10-07
author-reported LiveBench · Reason through hard problems	33	EXAONE 3.5 2.4B	model author	2026-10-07
author-reported LogicKor	8.51	EXAONE 3.5 2.4B	model author	2026-10-07
author-reported MT-Bench · Chat and follow instructions	7.81	EXAONE 3.5 2.4B	model author	2026-10-07
author-reported MT-Bench · Chat and follow instructions	7.24	EXAONE 3.5 2.4B	model author	2026-10-07

What the card says

3 cited sentences

Summary generated, cited

EXAONE-3.5-2.4B-Instruct is a 2.41B parameter bilingual instruction-tuned generative model developed by LG AI Research.^[1]^[2] The model supports English and Korean and features a context length of 32,768 tokens.^[3]^[4] It is released under the exaone license and is optimized for deployment on small or resource-constrained devices.^[5]^[6]

parameters human
“instruction-tuned bilingual (English and Korean) generati…”
languages
“Context Length: 32,768 tokens”
license
“optimized for deployment on small or resource-constrained…”

Generated 2026-10-07 15:13 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

3 use cases

What it is good at 3 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.

Chat and follow instructionsauthor-reportedAuthor reports MT-Bench 7.81 Author reports Arena-Hard 48.2 Author reports AlpacaEval 37.1 +5 more
Reason through hard problemsauthor-reportedAuthor reports LiveBench 33.0
Run on a laptop or phonestructuraldecision-model estimate p=1.00 fits: single-gpu

Facts

22 fields

Listed in LLMs

Facts

EXAONE-3.5-2.4B-Instruct has 2.41B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 32,768 tokens (32K). (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 2024-12-01, last modified 2026-02-06. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	ExaoneForCausalLM
Model type	exaone
Dtype	float32
Params by dtype	F32: 2,405,327,360
Size band	M
MoE	no
Context length	32,768 (config.json max_position_embeddings)
Languages	en, ko
Papers	arXiv:2412.04862
License	exaone (other)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · ExaoneForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · ExaoneForCausalLM · 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-3.5-2.4B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	5.8 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	2.9 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	1.4 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Original model Family: LGAI-EXAONE/EXAONE-3.5 120 derivatives on the Hub

finetune: 67quantized: 23adapter: 30

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
LGAI-EXAONE/EXAONE-Deep-2.4B	finetune	—	2.41B	M	other	LGAI-EXAONE	102	408.3k
lmstudio-community/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	lmstudio-community	2	83.3k
bartowski/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	bartowski	6	31k
mobilint/EXAONE-3.5-2.4B-Instruct	quantized	—	262M	S	other	mobilint	0	9.6k
Mungert/EXAONE-Deep-2.4B-GGUF	finetune	GGUF	—	—	other	Mungert	3	7.5k
QuantFactory/EXAONE-Deep-2.4B-GGUF	finetune	GGUF	—	—	other	QuantFactory	2	3.4k
second-state/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	second-state	0	3.4k
mobilint/EXAONE-3.5-2.4B-Instruct-Batch16	quantized	—	262M	S	other	mobilint	0	2.5k
MaziyarPanahi/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	—	MaziyarPanahi	2	2.4k
tensorblock/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	tensorblock	0	2.1k
mobilint/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	mobilint	0	2.1k
DevQuasar-8/LGAI-EXAONE.EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	—	DevQuasar-8	0	2k
gaianet/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	gaianet	0	1.7k
mobilint/EXAONE-3.5-2.4B-Instruct-regulus-rb-usb	quantized	—	262M	S	other	mobilint	0	1.2k
onnx-community/EXAONE-3.5-2.4B-Instruct	quantized	—	—	—	other	onnx-community	3	751
lunahr/thea-pro-2b-100r	finetune	—	2.41B	M	other	lunahr	3	511
huggingface-KREW/EXAGIRL-2.4B-Instruct	finetune	—	2.41B	M	other	huggingface-KREW	2	287
Edge-Quant/EXAONE-3.5-2.4B-Instruct-Q4_K_M-GGUF	quantized	GGUF	—	—	other	Edge-Quant	1	251
NikolayKozloff/EXAONE-3.5-2.4B-Instruct-Q8_0-GGUF	quantized	Q8_0	—	—	other	NikolayKozloff	1	235
FloatDo/EXAONE-3.5-2.4B-Instruct-GGUF	quantized	GGUF	—	—	other	FloatDo	0	272
alexgusevski/EXAONE-3.5-2.4B-Instruct-q8-mlx	quantized	mlx	2.41B	M	other	alexgusevski	0	230
alexgusevski/EXAONE-3.5-2.4B-Instruct-q4-mlx	quantized	mlx	2.41B	M	other	alexgusevski	0	206
alexgusevski/EXAONE-3.5-2.4B-Instruct-mlx	finetune	mlx	2.41B	M	other	alexgusevski	0	201
alexgusevski/EXAONE-3.5-2.4B-Instruct-q3-mlx	quantized	mlx	2.41B	M	other	alexgusevski	0	185
itlwas/EXAONE-3.5-2.4B-Instruct-Q4_K_M-GGUF	quantized	GGUF	—	—	other	itlwas	0	183
KCh3dRi4n/gongyil-exaone-24	finetune	—	1.28B	S	other	KCh3dRi4n	0	168
kyu5787/exaone-2.4b-mlx	finetune	mlx	2.41B	M	other	kyu5787	0	148
khj0816/EXAONE-Ian	adapter	—	—	—	—	khj0816	0	140
alexgusevski/EXAONE-3.5-2.4B-Instruct-q6-mlx	quantized	mlx	2.41B	M	other	alexgusevski	0	117
dlfp22/exaone-star-lora-best	adapter	—	—	—	—	dlfp22	0	116
nileshmalpeddi/output_exaone_cp_bylayer_5data_rank16_b4_g1_s2_e1_b512	finetune	—	2.61B	M	—	nileshmalpeddi	0	110
OpenLLM-Korea/EXAONE-Deep-2.4B	finetune	—	2.41B	M	other	OpenLLM-Korea	0	109
JunHaHwang/EXASPO-3.5-2.4B-Instruct	finetune	—	2.41B	M	other	JunHaHwang	0	108
jjkim110523/EXAONE-3.5-2.4B-Instruct-Q8_0-GGUF	quantized	Q8_0	—	—	other	jjkim110523	0	107
dlfp22/exaone-masking-lora-best	adapter	—	—	—	—	dlfp22	0	104
nileshmalpeddi/output_exaone_rank8_newdata_b4_g2_gradient2_e1_b1024	finetune	—	1.88B	S	—	nileshmalpeddi	0	101
Korshort/emnlp2026_expertqa_exaone3.5-2.4b_exp0_medical_fullft	finetune	—	2.41B	M	cc-by-nc-4.0	Korshort	0	85
Korshort/emnlp2026_expertqa_exaone3.5-2.4b_exp2_finance_fullft	finetune	—	2.41B	M	cc-by-nc-4.0	Korshort	0	85
space1637/iapyx-exaone-3.5-2.4b-sft	finetune	—	2.41B	M	other	space1637	0	83
mindw96/exaone-3.5-2.4b-instruct-DACON-LLM	finetune	—	2.67B	M	—	mindw96	0	82

Repository changelog

4 commits

Repository changelog 4 commits

Date	Commit	SHA
2026-02-06	Update README.md and config.json for Transformers v5	ccce25b
2026-02-04	Fix _tied_weights_keys mapping for Transformers v5 (#10)	c38726a
2024-12-11	Update README.md	e949c91
2024-12-09	Initial commit	8e6fc27

Articles and coverage

1 piece

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official1

Open-sourcing Three EXAONE 3.5 Models: 2.4B, 7.8B, 32B

lgresearch.ai2024-12-09officialabout the EXAONE 3.5 familyHN 13 pts · 4 comments1,244 words

LG AI Research released three EXAONE 3.5 models including a 2.4B, 7.8B, and 32B variant. All three models handle contexts of up to 32K tokens and ranked first in average scores for Instruction Following capabilities.generated, cited

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 3.5: Series of Large Language Models for Real-world Use Cases this model

Soyoung An, Kyunghoon Bae, Eunbi Choi +29 · 2024-12-06 · PDF · HF paper page (52 upvotes)

Similar models

3 models

Similar models

gemma-2-2b-it · 2.61B
same size band (M), same task (text-generation), more providers
gemma-2-2b · 2.61B
same size band (M), same task (text-generation), more providers
MiniCPM5-2B · 2.52B
same size band (M), same task (text-generation), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:41 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-3.5-2.4B-Instruct

Watch this model

/feeds/models/LGAI-EXAONE--EXAONE-3.5-2.4B-Instruct.xml(RSS 2.0) · every change to EXAONE-3.5-2.4B-Instruct
/feeds/families/LGAI-EXAONE--EXAONE-3.5.xml(RSS 2.0) · every change across the EXAONE-3.5 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.