Skip to content
Model Pareto

Kimi Linear 48B A3B Instruct

Moonshot AI (Kimi)Open weightsartificialanalysis.ai ↗
Input
—
Output
—
Speed
~100tok/sest.
Context
1Mtokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: Kimi Linear 48B A3B Instruct. 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× H200 141GB
FP8 weights
Throughput
~7,486 tok/s
many requests batched
Price
$0.051–0.42 /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)
  • FP8 weights (48B params, 3B active per token) + 40% KV-cache headroom ≈ 67 GB
  • 1× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~7,486 output tok/s aggregate at batch 256 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~155,846 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.

Writing & chat

Hard reasoning

Agents

OSWorld 2.0Estimated
~#3 of 5 (estimated)~53.8%
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. ~#3Kimi Linear 48B A3B InstructEstimated53.8%
  4. #3GPT-5.6 Terra50.2%
No published score — estimated from related benchmarks
~#61 of 92 (estimated)~636 Elo
  1. #1Claude Opus 5.51866
  2. #2Muse Wonder Full1857
  3. #60MiniCPM5-2B694
  4. ~#61Kimi Linear 48B A3B InstructEstimated636
  5. #61Nemotron 3.5 Lightning624
No published score — estimated from related benchmarks
~last of 91 (estimated)below measured range (<3.3%)
  1. #1Muse Spark 1.350.5%
  2. #2GLM-5.350.3%
  3. #90Llama 4 Scout3.3%
  4. ~#91Kimi Linear 48B A3B InstructEstimated<3.3%
No published score — estimated from related benchmarks