Skip to content
Model Pareto

Llama 3.2 Instruct 90B (Vision)

MetaOpen weightsartificialanalysis.ai ↗
Compare with nearest rivalsCompare with…
Input
—
Output
—
Speed
~63tok/sest.
Context
128Ktokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: Llama 3.2 Instruct 90B (Vision). 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× B200 180GB
FP8 weights
Throughput
~1,651 tok/s
many requests batched
Price
$0.47–4.23 /M tok
busy → light use
On your own machine
Mac Studio M5 Max 128GB (int4, ~10 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (88.6B params) + 25% KV-cache headroom ≈ 111 GB
  • 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~1,651 output tok/s aggregate at batch 47 (bandwidth-bound)
  • Blended 3:1 input:output; prefill ~11,999 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.

Vision

MMMUEstimated
~#2 of 3 (estimated)~71.3%
  1. #1Claude Haiku 4.573.2%
  2. ~#2Llama 3.2 Instruct 90B (Vision)Estimated71.3%
  3. #2Llama 4 Maverick69.4%
No published score — estimated from related benchmarks
CharXiv ReasoningEstimated
~last of 9 (estimated)below measured range (<74.5%)
  1. #1Gemini 3.8 Flash86.2%
  2. #2Kimi K384.8%
  3. #8Gemini 3.5 Flash-Lite74.5%
  4. ~#9Llama 3.2 Instruct 90B (Vision)Estimated<74.5%
No published score — estimated from related benchmarks
OmniDocBench v1.5Estimated
~last of 8 (estimated)below measured range (<75.8%)
  1. #1MiniMax-M391.6%
  2. #2Kimi K391.1%
  3. #7Muse Glimmer75.8%
  4. ~#8Llama 3.2 Instruct 90B (Vision)Estimated<75.8%
No published score — estimated from related benchmarks
GDP.pdf (AA)Estimated
~#41 of 43 (estimated)~0.2%
  1. #1GPT-6 Astra31.0%
  2. #2Muse Spark 1.326.6%
  3. #40Ministral 3 8B0.8%
  4. ~#41Llama 3.2 Instruct 90B (Vision)Estimated0.2%
  5. #41Ministral 3 3B0.0%
No published score — estimated from related benchmarks
ChartographyEstimated
~last of 23 (estimated)below measured range (<9.0%)
  1. #1GPT-6 Astra71.0%
  2. #2Claude Opus 5.566.3%
  3. #22Mistral Large 39.0%
  4. ~#23Llama 3.2 Instruct 90B (Vision)Estimated<9.0%
No published score — estimated from related benchmarks

Hard reasoning

AIME (latest)Estimated
~#20 of 25 (estimated)~27.2%
  1. #1Inkling97.1%
  2. #2Muse Glimmer94.7%
  3. #19Ministral 3 14B30.0%
  4. ~#20Llama 3.2 Instruct 90B (Vision)Estimated27.2%
  5. #20Ministral 3 3B22.0%
No published score — estimated from related benchmarks