Skip to content
Model Pareto

Llama 4 Maverick

MetaOpen weightsai.meta.com ↗
Input
$0.26/M tok
Output
$0.91/M tok
Blended
$0.422/M tok · 3:1
Speed
38tok/s
Context
1Mtokens
Cost per task
~$0.38/taskest.

Previous-generation open-weights model, for context. Price is the median third-party 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
8× H200 141GB
FP8 weights
Throughput
~25,249 tok/s
many requests batched
Price
$0.14–1.17 /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 (401.6B params, 17B active per token) + 40% KV-cache headroom ≈ 562 GB
  • 8× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~25,249 output tok/s aggregate at batch 908 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~220,018 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

~#23 of 24 (estimated)~44.4%
  1. #1Claude Fable 5.293.0%
  2. #2Claude Opus 5.589.9%
  3. #22Muse Glimmer51.2%
  4. ~#23Llama 4 MaverickEstimated44.4%
  5. #23Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#43 of 46 (estimated)~35.3%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #42Nex-N2.5-Mini36.1%
  4. ~#43Llama 4 MaverickEstimated35.3%
  5. #43Hy328.0%
No published score — estimated from related benchmarks
~#25 of 26 (estimated)~27.2
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #24GPT-6 Luna31.1
  4. ~#25Llama 4 MaverickEstimated27.2
  5. #25Muse Spark Open26.3
No published score — estimated from related benchmarks

Writing & chat

Vision

~last of 12 (estimated)below measured range (<74.5%)
  1. #1Gemini 3.8 Flash86.2%
  2. #2Kimi K384.8%
  3. #11Gemini 3.5 Flash-Lite74.5%
  4. ~#12Llama 4 MaverickEstimated<74.5%
No published score — estimated from related benchmarks
~last of 8 (estimated)below measured range (<75.8%)
  1. #1MiniMax-M391.6%
  2. #2Kimi K391.1%
  3. #7Muse Glimmer75.8%
  4. ~#8Llama 4 MaverickEstimated<75.8%
No published score — estimated from related benchmarks
GDP.pdf (AA)Estimated
~#41 of 48 (estimated)~2.4%
  1. #1Muse Wonder Full32.6%
  2. #2GPT-6 Astra31.0%
  3. #40Mistral Medium 3.52.8%
  4. ~#41Llama 4 MaverickEstimated2.4%
  5. #41Ministral 3 14B1.2%
No published score — estimated from related benchmarks
ChartographyEstimated
~last of 25 (estimated)below measured range (<9.0%)
  1. #1Claude Fable 5.273.4%
  2. #2GPT-6 Astra71.0%
  3. #24Mistral Large 39.0%
  4. ~#25Llama 4 MaverickEstimated<9.0%
No published score — estimated from related benchmarks

Hard reasoning

Agents