K-EXAONE 2.0 TPS calculator
Each card below is assessed against this model at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from the card's memory bandwidth and the size of the model once compressed.
Calculated for this model
818 cards we hold specifications for
Which GPUs can run K-EXAONE 2.0?
Set the inputs, read the answer
A longer conversation needs more memory, which can push this model off smaller cards.
Hides cards that would only fit the model by compressing it below this point.
0 cards match
Calculating| Needs | Quantisation | Fit | |||||
|---|---|---|---|---|---|---|---|
|
No card in our catalogue can run this model with these settings. |
|||||||
Speeds are estimates for a single request — one conversation at a time — calculated from memory bandwidth, model size and quantisation. Real throughput varies with the inference runtime and its version. Figures published by hardware vendors measure many simultaneous requests and are much higher.
On record
Full specification
Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.
Origin
Who built this model, where, and when it was published.
- Organisation
- LG AI Research
- Organisation type
- Industry
- Country
- Korea (Republic of)
- Published
- 31 July 2026
- Authors
- Eunbi Choi, Kibong Choi, Sehyun Chun, Seokhee Hong, Junwon Hwang, Hyojin Jeon, Ahra Jo, Hyunjik Jo, Yeonsik Jo, Minhyeok Jung, Doyoung Kim, Heegyu Kim, Joonkee Kim, Seonghwan Kim, Soyeon Kim, Sunkyoung Kim, Yireun Kim, Yongil Kim, Byungoh Ko, Changhun Lee, Dohaeng Lee, Haeju Lee, Jinsik Lee, Kyungmin Lee, Minwoo Lee, Wonkee Lee, Sangha Park, Sungjune Park, Kwangrok Ryoo, Kijung Seo, Minju Seo, Yon…
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Language modeling/generation, Chat, Question answering, Code generation, Mathematical reasoning, Instruction interpretation
- Base model
- K-EXAONE
Size
How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.
- Parameters
- 750B
- Training data
- tokens
MoE architecture: 750B total parameters, ~37B activated per token. 78 layers (2 dense + 76 MoE); 256 routed experts + 1 shared expert, top-8 routed experts activated per token. Upcycled from K-EXAONE (236B total / 23B active) by depth expansion (48 -> 78 layers) and expert duplication (128 -> 256 experts per layer). Excludes auxiliary speculative-decoding modules (MTP 0.52B, DSpark 2.53B).
16T tokens (developer-reported via email, July 2026). The technical report describes continual pre-training on an additional ~8T tokens after a healing stage, mid-training on 800B tokens (two 400B stages extending context to 64K then 256K), and SFT on 350B tokens, on top of the upcycled K-EXAONE base (11T tokens).
Training compute
The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.
- Training compute
- 3.6 × 10²⁴ FLOP
- How it was established
- Reported
3.55e24 FLOP reported by LG AI Research via email (July 2026), alongside a 16T-token training data figure. Cross-check by operation counting: 6 * 37e9 active parameters * 16e12 tokens = 3.55e24 FLOP, exactly consistent with the reported figure. Note that K-EXAONE 2.0 is upcycled from K-EXAONE (11T tokens at 23B active parameters, 1.52e24 FLOP). The technical report describes a healing stage followed by ~8T tokens of continual pre-training, 800B tokens of mid-training (2 x 400B), and 350B token…
Availability
Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.
- Weights
- Open — downloadable
- Model access
- Open weights (unrestricted)
- Training code
- Unreleased
- Hugging Face
- LGAI-EXAONE
Released under Apache 2.0 on Hugging Face.
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Why it is tracked
- Training cost
- Record confidence
- Confident
Developer-reported training compute cost of USD 17,998,771.54 (KRW 26,006,425,000 at 1,444.9 KRW/USD), communicated by LG AI Research via email (July 2026); exceeds the training cost threshold. Underlying cost calculation not provided.
Sources
Where this record came from and when it was last checked.
- Reference
- K-EXAONE 2.0 Technical Report
- Last updated
- 4 August 2026
What the numbers mean
What it takes to run this model
K-EXAONE 2.0 reaches a parameter count of 750B. That is beyond what any single graphics card holds. Running it means either splitting it across several cards or renting hardware built for the job, and every card able to hold it alone is a datacentre part. The number that can: 0.
Background
K-EXAONE 2.0 was published by LG AI Research, in the country recorded as Korea (Republic of), during July 2026. The publishing organisation is categorised as industry.
It works in the domain of Language, and is recorded as performing the task of language modeling/generation, Chat, Question answering, Code generation, Mathematical reasoning, Instruction interpretation.
Its starting point was an existing base model, K-EXAONE. Most models at this scale are adapted from an existing base rather than built from nothing.
Published weights mean the model runs on your machine rather than someone else's, which is what makes the hardware question below answerable at all. On Hugging Face it is published under the organisation LGAI-EXAONE.
How it was trained
Producing it required arithmetic totalling around 3.6 × 10²⁴ FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.
It is tracked in the underlying dataset for one reason in particular: training cost.
Step by step
How to choose a GPU for K-EXAONE 2.0
The table above has already assessed every card we hold specifications for against this model. Getting to your answer takes six steps.
-
01
Start from the memory column
The table lists every card able to hold K-EXAONE 2.0. No amount of processing power compensates for a card that cannot hold it.
-
02
Match the context to your actual use
The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect, because at long context a card that handles short questions easily can be dropped by K-EXAONE 2.0.
-
03
Set a quality floor
Each card runs the least-compressed copy it can hold. Setting a minimum quality drops the cards that only manage it by squeezing further than you would want, and holds the comparison at one level.
-
04
Sort by speed
Ranking by tokens per second follows memory bandwidth rather than core counts, for K-EXAONE 2.0. It will not match a gaming ordering, because generation is bound by memory bandwidth.
-
05
Read the fit column last
Tight means it loads and works with no room to raise the context later, in the case of K-EXAONE 2.0. Comfortable means you can grow the context later. That difference matters more than a few tokens per second, so buy for comfortable if you expect to.
-
06
Check the card from the other side
Every card name links to its own page, which runs the same calculation across the whole model catalogue. A card is usually bought for more than one model, so it is worth a look before buying for K-EXAONE 2.0.
Answers
K-EXAONE 2.0 — common questions
K-EXAONE 2.0— how many parameters does it have?
It has a parameter count of 750B. MoE architecture: 750B total parameters, ~37B activated per token. 78 layers (2 dense + 76 MoE); 256 routed experts + 1 shared expert, top-8 routed experts activated per token. Upcycled from K-EXAONE (236B total / 23B active) by depth expansion (48 -> 78 layers) and expert duplication (128 -> 256 experts per layer). Excludes auxiliary speculative-decoding modules (MTP 0.52B, DSpark 2.53B). That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.
K-EXAONE 2.0— who created it?
It was published by LG AI Research, based in Korea (Republic of), an organisation categorised as industry.
K-EXAONE 2.0— when was it released?
It was published in July 2026.
K-EXAONE 2.0— what is it used for?
It works in the domain of Language, and is recorded as handling the task of language modeling/generation, Chat, Question answering, Code generation, Mathematical reasoning, Instruction interpretation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
K-EXAONE 2.0— where can I download it?
Its weights are published on Hugging Face, under the organisation LGAI-EXAONE. We do not host model files — this site calculates what hardware is needed to run them.
K-EXAONE 2.0— how much compute was used to train it?
Training consumed around 3.6 × 10²⁴ FLOP. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.
K-EXAONE 2.0— can I run it if it does not fit in my GPU?
It can be split between the card and system memory, but it generates painfully slowly that way. The nearest miss we calculate falls short by 197.8 GB. Every figure here assumes the whole model is resident on the card.
K-EXAONE 2.0— would two GPUs run it faster?
Two cards buy memory rather than speed, which matters only if one card cannot hold it. The number that can: 0. So a second card is rarely the answer here.
K-EXAONE 2.0— why does the quantisation differ between cards?
Each card is shown running the least-compressed copy it can hold. The number of distinct compression levels across the cards that fit it: 1. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.
K-EXAONE 2.0— how accurate are these speed estimates?
Every figure is derived from memory bandwidth and model size, not benchmarked, which is why each is published as a range rather than a single number. One example: the range beneath each figure. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.
K-EXAONE 2.0— is it open source?
Its weights are published, so it can be downloaded and run on your own hardware. Note that open weights is not the same as open source in the full sense — it says nothing about the training data, the training code, or the commercial terms attached.
The other direction
Looking at it from the other side?
This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.