Skip to content

K2 Horizon 3.7B VRAM requirements

How much VRAM K2 Horizon 3.7B needs
4-bit weights: 2.9 GBWith 8K context: 5.0 GBWith 32K context: 8.4 GB8-bit, 8K context: 7.1 GB
Parameters
5.1B
Longest context
512K
Released
Sep 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM K2 Horizon 3.7B needs

Memory K2 Horizon 3.7B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit9.4 GB11.6 GB14.9 GB28.4 GB
8-bit5.0 GB7.1 GB10.5 GB24.0 GB
6-bit3.9 GB6.0 GB9.4 GB22.9 GB
5-bit3.4 GB5.5 GB8.9 GB22.4 GB
4-bit2.9 GB5.0 GB8.4 GB21.9 GB
3-bit2.3 GB4.4 GB7.8 GB21.3 GB
2-bit1.8 GB3.9 GB7.3 GB20.8 GB
1.6-bit1.0 GB3.1 GB6.5 GB20.0 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run K2 Horizon 3.7B

The best way K2 Horizon 3.7B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40606-bit16K50 tok/s0.6 sPDF: K2 Horizon 3.7B on RTX 4060 report
RTX 50606-bit16K79 tok/s0.4 sPDF: K2 Horizon 3.7B on RTX 5060 report
RTX 30706-bit16K79 tok/s0.4 sPDF: K2 Horizon 3.7B on RTX 3070 report
RTX 3080 10 GB8-bit24K101 tok/s0.3 sPDF: K2 Horizon 3.7B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit36K51 tok/s0.7 sPDF: K2 Horizon 3.7B on RTX 3060 12 GB report
Arc B5808-bit36K44 tok/s0.5 sPDF: K2 Horizon 3.7B on Arc B580 report
RTX 40708-bit36K70 tok/s0.3 sPDF: K2 Horizon 3.7B on RTX 4070 report
RTX 4070 Super8-bit36K70 tok/s0.2 sPDF: K2 Horizon 3.7B on RTX 4070 Super report
RTX 50708-bit36K90 tok/s0.3 sPDF: K2 Horizon 3.7B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit64K41 tok/s0.4 sPDF: K2 Horizon 3.7B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit64K62 tok/s0.4 sPDF: K2 Horizon 3.7B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit64K53 tok/s0.4 sPDF: K2 Horizon 3.7B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit64K71 tok/s0.2 sPDF: K2 Horizon 3.7B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit64K90 tok/s0.2 sPDF: K2 Horizon 3.7B on RTX 4070 Ti Super report
RTX 4080 Super8-bit64K98 tok/s0.2 sPDF: K2 Horizon 3.7B on RTX 4080 Super report
RTX 5070 Ti8-bit64K117 tok/s0.2 sPDF: K2 Horizon 3.7B on RTX 5070 Ti report
RTX 50808-bit64K124 tok/s0.2 sPDF: K2 Horizon 3.7B on RTX 5080 report
Radeon RX 7900 XTX8-bit112K101 tok/s0.4 sPDF: K2 Horizon 3.7B on Radeon RX 7900 XTX report
RTX 30908-bit112K121 tok/s0.2 sPDF: K2 Horizon 3.7B on RTX 3090 report
RTX 40908-bit112K129 tok/s0.1 sPDF: K2 Horizon 3.7B on RTX 4090 report
Radeon AI PRO R97008-bit168K71 tok/s0.2 sPDF: K2 Horizon 3.7B on Radeon AI PRO R9700 report
RTX 50908-bit168K207 tok/s0.1 sPDF: K2 Horizon 3.7B on RTX 5090 report
2× RTX 30908-bit256K121 tok/s0.1 sPDF: K2 Horizon 3.7B on 2× RTX 3090 report
2× RTX 40908-bit256K129 tok/s0.1 sPDF: K2 Horizon 3.7B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K165 tok/s0.1 sPDF: K2 Horizon 3.7B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K124 tok/s0.1 sPDF: K2 Horizon 3.7B on RTX 6000 Ada report
L40S8-bit256K113 tok/s0.1 sPDF: K2 Horizon 3.7B on L40S report
2× RTX 50908-bit384K207 tok/s0.1 sPDF: K2 Horizon 3.7B on 2× RTX 5090 report
4× RTX 30908-bit512K121 tok/s0.1 sPDF: K2 Horizon 3.7B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit512K207 tok/s0.1 sPDF: K2 Horizon 3.7B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit28K20 tok/s5.4 sPDF: K2 Horizon 3.7B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit64K20 tok/s5.4 sPDF: K2 Horizon 3.7B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit96K20 tok/s5.4 sPDF: K2 Horizon 3.7B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit64K42 tok/s2.7 sPDF: K2 Horizon 3.7B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit200K42 tok/s2.7 sPDF: K2 Horizon 3.7B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K42 tok/s2.7 sPDF: K2 Horizon 3.7B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit512K70 tok/s1.4 sPDF: K2 Horizon 3.7B on M4 Max 128 GB report
M5 Max 128 GB8-bit512K77 tok/s0.4 sPDF: K2 Horizon 3.7B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit416K86 tok/s0.8 sPDF: K2 Horizon 3.7B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit512K86 tok/s0.8 sPDF: K2 Horizon 3.7B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit512K86 tok/s0.8 sPDF: K2 Horizon 3.7B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit512K112 tok/s0.2 sPDF: K2 Horizon 3.7B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K31 tok/s0.8 sPDF: K2 Horizon 3.7B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit512K31 tok/s0.8 sPDF: K2 Horizon 3.7B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit512K43 tok/s0.3 sPDF: K2 Horizon 3.7B on RTX Spark report
DGX Spark8-bit512K39 tok/s0.3 sPDF: K2 Horizon 3.7B on DGX Spark report
Dual DGX Spark8-bit512K39 tok/s0.2 sPDF: K2 Horizon 3.7B on Dual DGX Spark report
DGX Station GB3008-bit512K494 tok/s0.1 sPDF: K2 Horizon 3.7B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit512K207 tok/s0.1 sPDF: K2 Horizon 3.7B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit512K207 tok/s0.1 sPDF: K2 Horizon 3.7B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit512K207 tok/s0.1 sPDF: K2 Horizon 3.7B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit480K229 tok/s0.2 sPDF: K2 Horizon 3.7B on A100 80 GB SXM report
H100 SXM8-bit480K325 tok/s0.1 sPDF: K2 Horizon 3.7B on H100 SXM report
4× H200 NVL8-bit512K404 tok/s0.1 sPDF: K2 Horizon 3.7B on 4× H200 NVL report
8× H200 NVL8-bit512K404 tok/s0.1 sPDF: K2 Horizon 3.7B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf IFM/K2-Horizon-3.7B-GGUF
vLLM
vllm serve IFM/K2-Horizon-3.7B
MLX (Mac)
mlx_lm.chat --model mlx-community/K2-Horizon-3.7B-4bit

Questions

How much VRAM does K2 Horizon 3.7B need?

K2 Horizon 3.7B needs 2.9 GB for its weights at 4-bit, 5.0 GB with an 8K context and 8.4 GB with 32K. At 8-bit, 7.1 GB.

What is the minimum VRAM for K2 Horizon 3.7B?

About 4.4 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is K2 Horizon 3.7B?

5.1B parameters: 9.4 GB of weights at 16-bit, 2.9 GB at 4-bit.

Can K2 Horizon 3.7B run on an RTX 4090?

Yes. At 8-bit it holds 112K of context and writes about 129 tokens a second.

More models