Skip to content
Model Pareto

Qwen3.5 397B A17B

Alibaba QwenOpen weightsartificialanalysis.ai ↗
Input
$0.60/M tok
Output
$3.60/M tok
Blended
$1.35/M tok · 3:1
Speed
86tok/s
Context
262Ktokens
Cost per task
~$1.5/taskest.

Long-tail import from Artificial Analysis (2026-09-24); AA variant: Qwen3.5 397B A17B (Reasoning). Price = AA median host price.

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
4× B200 180GB
FP8 weights
Throughput
~12,969 tok/s
many requests batched
Price
$0.20–1.76 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (int4, ~99 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (397B params, 17B active per token) + 25% KV-cache headroom ≈ 496 GB
  • 4× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~12,969 output tok/s aggregate at batch 371 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~250,147 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

~#13 of 21 (estimated)~73.7%
  1. #1Claude Opus 596.0%
  2. #2Claude Sonnet 585.2%
  3. #12Flint74.2%
  4. ~#13Qwen3.5 397B A17BEstimated73.7%
  5. #13Claude Haiku 4.573.3%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#15 of 28 (estimated)~68.3%
  1. #1Qwen3.8-Omni-Flash92.6%
  2. #2Qwen3.8 Flash-Next91.9%
  3. #14Maple-Preview75.1%
  4. ~#15Qwen3.5 397B A17BEstimated68.3%
  5. #15Mistral Small 464.0%
No published score — estimated from related benchmarks
~#24 of 26 (estimated)~32.6
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #23Gemini 3.8 Flash33.3
  4. ~#24Qwen3.5 397B A17BEstimated32.6
  5. #24GPT-6 Luna31.1
No published score — estimated from related benchmarks

Writing & chat

Vision

Hard reasoning

Agents