Skip to content
Model Pareto

K2 Horizon 375B A23B

MBZUAI Institute of Foundation ModelsOpen weightsartificialanalysis.ai ↗
Compare with nearest rivalsCompare with…
Input
—
Output
—
Speed
~88tok/sest.
Context
524Ktokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: K2 Horizon 375B A23B. 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
8× H200 141GB
FP8 weights
Throughput
~22,644 tok/s
many requests batched
Price
$0.17–1.38 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (int4, ~73 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (375B params, 23B active per token) + 40% KV-cache headroom ≈ 525 GB
  • 8× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~22,644 output tok/s aggregate at batch 814 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~162,622 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

SWE-bench VerifiedEstimated
~#10 of 14 (estimated)~74.8%
  1. #1Claude Opus 596.0%
  2. #2Claude Sonnet 585.2%
  3. #9Xing4.0-29B-A4B75.0%
  4. ~#10K2 Horizon 375B A23BEstimated74.8%
  5. #10Claude Haiku 4.573.3%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#28 of 31 (estimated)~50.6%
  1. #1DeepSeek V4.1 Flash74.2%
  2. #2Grok 4.772.6%
  3. #27Qwen3.8 Max (0902)51.0%
  4. ~#28K2 Horizon 375B A23BEstimated50.6%
  5. #28Qwen3.8 27B42.2%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#8 of 20 (estimated)~73.1%
  1. #1Qwen3.8-Omni-Flash92.6%
  2. #2Qwen3.8 Flash-Next91.9%
  3. #7gpt-oss-20b77.7%
  4. ~#8K2 Horizon 375B A23BEstimated73.1%
  5. #8Mistral Small 464.0%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~last of 15 (estimated)below measured range (<41.1)
  1. #1Claude Opus 5.566.0
  2. #2Claude Fable 5.162.2
  3. #14GPT-6 Luna41.1
  4. ~#15K2 Horizon 375B A23BEstimated<41.1
No published score — estimated from related benchmarks

Hard reasoning

AIME (latest)Estimated
~#3 of 25 (estimated)~94.6%
  1. #1Inkling97.1%
  2. #2Muse Glimmer94.7%
  3. ~#3K2 Horizon 375B A23BEstimated94.6%
  4. #3Nova 2.0 Lite94.3%
No published score — estimated from related benchmarks

Agents

τ²-bench (Telecom)Estimated
~#24 of 116 (estimated)~83.8%
  1. #1Step 3.7 Flash98.5%
  2. #2Gemini 3.1 Pro95.6%
  3. #23GPT-5.3 Codex86.0%
  4. ~#24K2 Horizon 375B A23BEstimated83.8%
  5. #24GPT-5.4 mini83.3%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#8 of 12 (estimated)~75.0%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #7GPT-5.578.7%
  4. ~#8K2 Horizon 375B A23BEstimated75.0%
  5. #8Gemini 3.5 Flash-Lite74.0%
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 375B A23BEstimated<47.5%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#14 of 18 (estimated)~77.9%
  1. #1Atria Dawn Preview92.5%
  2. #2GPT-6 Astra91.5%
  3. #13Nex-N2.5-Mini83.4%
  4. ~#14K2 Horizon 375B A23BEstimated77.9%
  5. #14Inkling77.1%
No published score — estimated from related benchmarks