Skip to content
Model Pareto

K2 Horizon 3.7B

MBZUAI Institute of Foundation ModelsOpen weightsartificialanalysis.ai ↗
Input
—
Output
—
Speed
~120tok/sest.
Context
524Ktokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: K2 Horizon 3.7B. Price = AA median host price.

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.

Hardware
1× B200 180GB
BF16 weights
Throughput
~13,869 tok/s
many requests batched
Price
$0.056–0.50 /M tok
busy → light use
On your own machine
RTX 5090 32GB (int4, 300+ tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • BF16 weights (5.06B params) + 40% KV-cache headroom ≈ 14 GB
  • 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~13,869 output tok/s aggregate at batch 256 (bandwidth-bound)
  • Blended 3:1 input:output; prefill ~105,052 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Coding

~#23 of 24 (estimated)~45.5%
  1. #1Claude Fable 5.293.0%
  2. #2Claude Opus 5.589.9%
  3. #22Muse Glimmer51.2%
  4. ~#23K2 Horizon 3.7BEstimated45.5%
  5. #23Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#43 of 46 (estimated)~33.4%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #42Nex-N2.5-Mini36.1%
  4. ~#43K2 Horizon 3.7BEstimated33.4%
  5. #43Hy328.0%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#21 of 28 (estimated)~45.8%
  1. #1Qwen3.8-Omni-Flash92.6%
  2. #2Qwen3.8 Flash-Next91.9%
  3. #20Mistral Large 346.5%
  4. ~#21K2 Horizon 3.7BEstimated45.8%
  5. #21Devstral 244.8%
No published score — estimated from related benchmarks
~#25 of 26 (estimated)~27.9
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #24GPT-6 Luna31.1
  4. ~#25K2 Horizon 3.7BEstimated27.9
  5. #25Muse Spark Open26.3
No published score — estimated from related benchmarks

Hard reasoning

Agents

~#12 of 15 (estimated)~64.5%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #11Muse Glimmer65.9%
  4. ~#12K2 Horizon 3.7BEstimated64.5%
  5. #12Nemotron 3.5 Super VL NVFP452.3%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. #4GPT-5.547.5%
  4. ~#5K2 Horizon 3.7BEstimated<47.5%
No published score — estimated from related benchmarks