How much VRAM does K2 Horizon 375B-A23B need?

About 249.7 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 257.0 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.

Will K2 Horizon 375B-A23B run on a 512GB GPU?

Yes — at Q4_K_M and 4K context it needs about 249.7 GB, which leaves 134.3 GB spare on a Mac M5 Ultra 512G. That is the smallest card in this index that clears it comfortably; 1 of 78 do.

Which quantization of K2 Horizon 375B-A23B should I use?

No published quality comparison exists for this model, so pick by footprint: Q4_K_M is the level most people run, and a higher bits-per-weight level is more faithful. All 4 levels in the index are Q2_K, Q3_K_M, Q4_K_M, Q8_0. Note this is a mixture-of-experts model — all parameters must be resident even though only a fraction are active per token, so the memory cost follows the total, not the active count.

K2 Horizon 375B-A23B

375B-A23B

MBZUAI K2 Horizon

The largest K2 Horizon model from MBZUAI IFM, Apache 2.0 with its training data: 379B parameters, about 23B active per token. The parameter count is taken from its configuration. Every one of its 61 layers keeps an ordinary cache. Q2_K (bartowski's 146 GB in four parts) needs about 154 GB at 4K, comfortable on a 256 GB Mac Studio; Q4_K_M (238 GB in seven parts) needs about 250 GB and is comfortable only on a 512 GB one. IFM publishes no GGUF for this size; the sizes here are bartowski's files, and no per-level quality loss has been published.

⬇ 2.5K HF downloads♥ 1 likesbartowski/K2-Horizon-375B-A23B-GGUF· stats from 10/23/2026

Mac / Apple SiliconPro GPU

524K

Max Context

Quant Variants

GGUF Q8_0

Best Quality

—

Perplexity rise (best build)

Calculate VRAM Hugging Face Compare

Quantization Variants

Per-quant VRAM, quality loss, and inference speed on RTX 4090

Measured = run on this site · Estimated = not run here: calculated, or a figure this site has not verified · Community = public reports

Format	Level	BPW	VRAM	PPL Loss	Speed	Source	Actions
GGUF	Q2_K	3.09	154.1 GB	—	—	Estimated	Calc HF
GGUF	Q3_K_M	3.85	191.7 GB	—	—	Estimated	Calc HF
GGUF	Q4_K_M	5.02	249.7 GB	—	—	Estimated	Calc HF
GGUF	Q8_0	8.5	422.0 GB	—	—	Estimated	Calc HF

Which file each size is for

A file size is not a memory figure: the VRAM column is the calculator's total at 4K context — weights plus KV cache plus a 10% activation buffer. Units differ on purpose: file sizes are decimal GB, as Hugging Face lists them; the VRAM column is GiB (1024³ bytes) with 2% added to the weights, so 7.12 GB of file counts as about 6.8 GiB there. A vision projector (mmproj), where the repo ships one, is a separate download and is in neither number.

GGUF Q2_Kbartowski/K2-Horizon-375B-A23B-GGUF · K2-Horizon-375B-A23B-Q2_K-00001-of-00004.gguf · 146.25 GB (146,249,929,920 bytes) · 4 parts · listing read 2026-10-23
GGUF Q3_K_Mbartowski/K2-Horizon-375B-A23B-GGUF · K2-Horizon-375B-A23B-Q3_K_M-00001-of-00005.gguf · 182.54 GB (182,537,573,664 bytes) · 5 parts · listing read 2026-10-23
GGUF Q4_K_Mbartowski/K2-Horizon-375B-A23B-GGUF · K2-Horizon-375B-A23B-Q4_K_M-00001-of-00007.gguf · 237.80 GB (237,797,004,832 bytes) · 7 parts · listing read 2026-10-23
GGUF Q8_0bartowski/K2-Horizon-375B-A23B-GGUF · K2-Horizon-375B-A23B-Q8_0-00001-of-00011.gguf · 403.08 GB (403,079,845,472 bytes) · 11 parts · listing read 2026-10-23

Similar models

Compare with K2 Horizon

36B-A4B

K2 Horizon MoVA 36B-A4B

MBZUAI K2 Horizon

Consumer GPUPro GPU

24.2 GBmin VRAM—perplexity rise

The mixture-of-experts member of MBZUAI IFM's K2 Horizon family: 37.5B parameters with about 4B active per token, Apache 2.0, released with its training data. Every one of its 48 layers keeps an ordinary KV cache, so context is not cheap: Q4_K_M is about 24.2 GB at 4K, which is tight on a 24 GB card, and comfortable on a 32 GB card up to 16K. The 512K window would need about 129 GB. llama.cpp runs it as `k2-horizon`. Sizes come from IFM's own GGUF files (bartowski's Q4_K_M is about 0.3 GB larger); no per-level quality loss has been published.

32B

K2 Horizon 32B

MBZUAI K2 Horizon

Consumer GPUPro GPU

23.1 GBmin VRAM—perplexity rise

The dense 32B member of MBZUAI IFM's K2 Horizon family (34.8B counted, Apache 2.0, released with its training data). All 64 layers keep an ordinary KV cache with 8 KV heads: Q4_K_M is about 23.1 GB at 4K, tight on a 24 GB card and comfortable on a 32 GB card up to 16K; each extra 32K of context adds roughly 8.8 GB. llama.cpp runs it as `k2-horizon`. Sizes come from IFM's own GGUF files (bartowski's Q4_K_M is about 1 GB larger); no per-level quality loss has been published.

K2 Horizon 7B

MBZUAI K2 Horizon

Consumer GPUMac / Apple Silicon

6.5 GBmin VRAM—perplexity rise

The small dense member of MBZUAI IFM's K2 Horizon family, Apache 2.0, released with its training data. It is called 7B but holds 9.0B parameters: the 250K-token vocabulary's input and output tables add about 2B. Q4_K_M is about 6.5 GB at 4K, comfortable on an 8 GB card at that length; a 16 GB card holds it comfortably to 32K (10.8 GB), and 128K needs about 25.6 GB because every layer keeps an ordinary KV cache. llama.cpp runs it as `k2-horizon`. Sizes come from IFM's own GGUF files (bartowski's Q4_K_M is about 0.2 GB larger); no per-level quality loss has been published.

0.9B

K2 Horizon 0.9B

MBZUAI K2 Horizon

Consumer GPUMac / Apple Silicon

1.0 GBmin VRAM—perplexity rise

The smallest member of MBZUAI IFM's K2 Horizon family, Apache 2.0 and released with its training data. It is called 0.9B and holds 1.08B parameters. At Q4_K_M it needs about 1.0 GB at 4K and 2.6 GB at 32K, so it runs on almost anything, including a CPU. Every layer keeps an ordinary cache, which is why its full 128K window needs about 8.4 GB. IFM's own GGUF repo names these files 1B; the sizes here are bartowski's files, and no per-level quality loss has been published.

Running K2 Horizon 375B-A23B locally

At Q4_K_M and 4K of context, K2 Horizon 375B-A23B needs about 249.7 GB — 226.0 GB of weights, 0.95 GB of KV cache and a 22.7 GB activation buffer. The smallest card in this index that clears that comfortably is the Mac M5 Ultra 512G at 512 GB, and 1 of the 78 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs.

What longer context costs

Going from 4K to 32K adds about 6.68 GB, taking the total to 257.0 GB. Weights do not move with context — only the KV cache does, and it grows linearly, so this is the number to watch when planning for long documents. The model's native window is 512K; holding all of it at Q4_K_M would need about 383 GB.

Which build to download

This index only tracks K2 Horizon 375B-A23B in GGUF, across 4 levels (Q2_K, Q3_K_M, Q4_K_M, Q8_0). No per-level perplexity sweep has been published for this model, so the quality column is empty rather than estimated — within one model, more bits per weight is the only ordering the data supports. No throughput has been measured for this model on this site's hardware. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 — CUDA only, and now archived, so local use rather than serving.

Common questions

How much VRAM does K2 Horizon 375B-A23B need?: About 249.7 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 257.0 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.
Will K2 Horizon 375B-A23B run on a 512GB GPU?: Yes — at Q4_K_M and 4K context it needs about 249.7 GB, which leaves 134.3 GB spare on a Mac M5 Ultra 512G. That is the smallest card in this index that clears it comfortably; 1 of 78 do.
Which quantization of K2 Horizon 375B-A23B should I use?: No published quality comparison exists for this model, so pick by footprint: Q4_K_M is the level most people run, and a higher bits-per-weight level is more faithful. All 4 levels in the index are Q2_K, Q3_K_M, Q4_K_M, Q8_0. Note this is a mixture-of-experts model — all parameters must be resident even though only a fraction are active per token, so the memory cost follows the total, not the active count.

Where this model fits

Sized at Q4_K_M with a 4K context window, smallest card first. Comfortable means the estimate uses at most 88% of the memory.

Runs comfortably on: Mac M5 Ultra 512G512GB · +134.3
Just misses: Mac M5 Ultra 256G+57.7GB over Mac M2 Ultra 192G+105.7GB over
Ships in: GGUF
Try it yourself: Size it in the calculator Generate the run command

How to actually run this

Deployment guides for this model and this class of hardware.

This page's figures change when the model or the runtime does.Last updated 2026-10-23 RSS → /feed.xml