K2 Horizon 375B-A23B is MBZUAI IFM's flagship: a 375B MoE with 23B active and a 512K native context, released with its training data, code and intermediate checkpoints. It reasons by default and is tuned for agentic tool use.
chatcodingreasoningmathagentictool_use
379.17B
Parameters (23B active)
512K
Context length
6
Benchmarks
17
Quantizations
15K
HF downloads
Architecture
MoE
Released
2026-09-03
Layers
61
KV Heads
8
Head Dim
128
Family
other
Quantization Options
Context length:
Quant
Bits
VRAM @ 16K
Quality
IQ2_XXS
2.38
116.2 GB
113.3 + 2.9 KV
low
IQ2_M
2.93
142.2 GB
139.4 + 2.9 KV
low
Q2_K
3.16
153.1 GB
150.3 + 2.9 KV
low
IQ3_XXS
3.25
157.4 GB
154.5 + 2.9 KV
low
IQ3_XS
3.5
169.2 GB
166.4 + 2.9 KV
low
Q3_K_S
3.64
175.9 GB
173.0 + 2.9 KV
low
IQ3_M
3.76
181.6 GB
178.7 + 2.9 KV
low
Q3_K_M
4
192.9 GB
190.1 + 2.9 KV
low
Q3_K_L
4.3
207.2 GB
204.3 + 2.9 KV
moderate
IQ4_XS
4.46
214.7 GB
211.9 + 2.9 KV
moderate
Q4_K_S
4.67
224.7 GB
221.8 + 2.9 KV
moderate
Q4_K_M
4.89
235.1 GB
232.3 + 2.9 KV
good
Q5_K_S
5.57
267.3 GB
264.5 + 2.9 KV
good
Q5_K_M
5.7
273.5 GB
270.6 + 2.9 KV
good
Q6_K
6.56
314.3 GB
311.4 + 2.9 KV
excellent
Q8_0
8.5
406.2 GB
403.4 + 2.9 KV
lossless
FP16
16
761.7 GB
758.8 + 2.9 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.