Skip to content
Model Pareto

Kimi K3

Moonshot AI (Kimi)Open weightskimi.com ↗
Compare with nearest rivalsCompare with…
Input
$3.00/M tok
Output
$15/M tok
Blended
$6.00/M tok · 3:1
Speed
37tok/s
Context
1Mtokens

Open weights (Kimi K3 License). Card: "Quantization: MXFP4 weights / MXFP8 activations (quantization-aware training)".

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.

Hardware
32× B200 180GB
FP8 weights
Throughput
~42,356 tok/s
many requests batched
Price
$0.61–5.44 /M tok
busy → light use
On your own machine
Too large
needs data-center GPUs
Assumptions (5)
  • FP8 weights (2800B params, 104B active per token) + 40% KV-cache headroom ≈ 3920 GB
  • 32× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~42,356 output tok/s aggregate at batch 1426 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~278,048 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
~#2 of 14 (estimated)~85.3%
  1. #1Claude Opus 596.0%
  2. ~#2Kimi K3Estimated85.3%
  3. #2Claude Sonnet 585.2%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#6 of 19 (estimated)~66.6%
  1. #1Claude Opus 5.589.9%
  2. #2Claude Sonnet 5.581.3%
  3. #5Intern-S2-397B68.5%
  4. ~#6Kimi K3Estimated66.6%
  5. #6Hy4 preview65.7%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#3 of 20 (estimated)~90.7%
  1. #1Qwen3.8-Omni-Flash92.6%
  2. #2Qwen3.8 Flash-Next91.9%
  3. ~#3Kimi K3Estimated90.7%
  4. #3Qwen3.8 27B90.3%
No published score — estimated from related benchmarks

Writing & chat

IFBenchEstimated
~#3 of 126 (estimated)~82.6%
  1. #1MiniMax-M382.9%
  2. #2Qwen3.8 2.4T A95B82.8%
  3. ~#3Kimi K3Estimated82.6%
  4. #3Qwen3.8-Omni-Flash81.5%
No published score — estimated from related benchmarks
MMMLUEstimated
~#2 of 5 (estimated)~92.0%
  1. #1Gemini 3.1 Pro92.6%
  2. ~#2Kimi K3Estimated92.0%
  3. #2Gemma 4 31B88.4%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#7 of 531477 Elo
  1. #1Claude Opus 5.51501
  2. #2Claude Fable 5.1Few votes1496
  3. #6Gemini 3.1 Pro1480
  4. #7Kimi K31477
  5. #8GPT-5.51474
lmarena.ai · 2026-09-25
LMArena Text (Elo)Verified
#6 of 531488 Elo
  1. #1Claude Opus 5.51509
  2. #2Claude Fable 5.11501
  3. #3Muse Spark 1.31493
  4. #6Kimi K31488
  5. #7Gemini 3.1 Pro1487
lmarena.ai · 2026-09-25

Vision

MMMUEstimated
~#2 of 3 (estimated)~71.3%
  1. #1Claude Haiku 4.573.2%
  2. ~#2Kimi K3Estimated71.3%
  3. #2Llama 4 Maverick69.4%
No published score — estimated from related benchmarks

Hard reasoning

AIME (latest)Estimated
~#2 of 25 (estimated)~96.8%
  1. #1Inkling97.1%
  2. ~#2Kimi K3Estimated96.8%
  3. #2Muse Glimmer94.7%
No published score — estimated from related benchmarks

Agents

τ²-bench (Telecom)Estimated
~#2 of 116 (estimated)~98.3%
  1. #1Step 3.7 Flash98.5%
  2. ~#2Kimi K3Estimated98.3%
  3. #2Gemini 3.1 Pro95.6%
No published score — estimated from related benchmarks