K2-Horizon-32B
- Input
- —
- Output
- —
- Speed
- ~92tok/sest.
- Context
- 524Ktokens
Open weights (Apache 2.0), no first-party API; plotted at self-host estimate. Card benchmark column is labelled K2-Horizon-32B-Stage1; the repo is the only 32B release and the card says all reported results use high reasoning effort.
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
- FP8 weights
- Throughput
- ~4,595 tok/s
- many requests batched
- Price
- $0.17–1.56 /M tok
- busy → light use
- On your own machine
- RTX 5090 32GB (int4, ~72 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (34.78B params) + 40% KV-cache headroom ≈ 49 GB
- 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~4,595 output tok/s aggregate at batch 131 (bandwidth-bound)
- Blended 3:1 input:output; prefill ~30,567 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
~#19 of 21 (estimated)~61.6%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #18gpt-oss-120b62.4%
- ~#19K2-Horizon-32BEstimated61.6%
- #19NVIDIA Nemotron-Labs Dory1.0 4B59.5%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~last of 24 (estimated)below measured range (<43.8%)
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#44 of 46 (estimated)~25.7%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #43Hy328.0%
- ~#44K2-Horizon-32BEstimated25.7%
- #44Nemotron 3.5 Super VL24.6%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Estimated
~#42 of 102 (estimated)~4.1%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #41Qwen3.8 27B5.6%
- ~#42K2-Horizon-32BEstimated4.1%
- #42Gemini 3.1 Pro4.0%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#17 of 28 (estimated)~55.1%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #16Claude Haiku 4.561.5%
- ~#17K2-Horizon-32BEstimated55.1%
- #17GLM-5.3 NVFP453.5%
No published score — estimated from related benchmarks
SciCodeLab-reported
#105 of 11830.2%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #104NVIDIA Nemotron 3 Nano 30B A3B30.6%
- #105K2-Horizon-32B30.2%
- #106K2 Horizon 7B27.5%
huggingface.co · 2026-09-01
AA Coding Agent IndexEstimated
~#25 of 26 (estimated)~28.1
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #24GPT-6 Luna31.1
- ~#25K2-Horizon-32BEstimated28.1
- #25Muse Spark Open26.3
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondLab-reported
#70 of 19282.3%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #70GPT-5.5 Instant (June 2026)82.3%
- #70K2-Horizon-32B82.3%
- #72K2 Horizon MoVA 36B A4B82.2%
huggingface.co · 2026-09-01
Humanity's Last ExamLab-reported
#71 of 19822.8%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #70K2 Horizon MoVA 36B A4B23.4%
- #71K2-Horizon-32B22.8%
- #72Ling-3.0-flash-Fin22.6%
huggingface.co · 2026-09-01
AIME (latest)Estimated
~#15 of 34 (estimated)~87.6%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #14Nova 2.0 Pro Preview89.0%
- ~#15K2-Horizon-32BEstimated87.6%
- #15Maple-Preview87.5%
No published score — estimated from related benchmarks
AA Intelligence IndexEstimated
~#55 of 103 (estimated)~22.9
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #54Solar Mini 424.0
- ~#55K2-Horizon-32BEstimated22.9
- #55Nemotron 3 Ultra22.9
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Estimated
~#41 of 120 (estimated)~72.8%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #39Nemotron 3.5 Super VL NVFP473.2%
- ~#41K2-Horizon-32BEstimated72.8%
- #41Nova 2.0 Lite72.8%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#11 of 15 (estimated)~67.9%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #10Nex-N2.5-Mini71.2%
- ~#11K2-Horizon-32BEstimated67.9%
- #11Muse Glimmer65.9%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
No published score — estimated from related benchmarks
BrowseCompEstimated
~#19 of 22 (estimated)~68.9%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #18Nemotron 3.5 Bolt 74B-A9B70.0%
- ~#19K2-Horizon-32BEstimated68.9%
- #19Mistral Medium 3.548.6%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Estimated
~#30 of 92 (estimated)~1359 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #29K2 Horizon 375B A23B1361
- ~#30K2-Horizon-32BEstimated1359
- #30GPT-5.51336
No published score — estimated from related benchmarks
AutomationBench-AAEstimated
~#39 of 85 (estimated)~23.0%
- #1Claude Fable 5.276.4%
- #2Muse Wonder Full71.3%
- #38Quasar 438B23.9%
- ~#39K2-Horizon-32BEstimated23.0%
- #39MiniMax-M321.3%
No published score — estimated from related benchmarks