Skip to content
Model Pareto

Mistral Large 3

Mistral AIOpen weightsmistral.ai ↗
Input
$0.50/M tok
Output
$1.50/M tok
Blended
$0.75/M tok · 3:1
Speed
77tok/s
Context
256Ktokens
Cost per task
~$0.63/taskest.

Previous-generation open-weights flagship. Main release is FP8 (card: "FP8 on a single node of B200s or H200s"; BF16 provided "if needed").

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× B200 180GB
FP8 weights
Throughput
~20,662 tok/s
many requests batched
Price
$0.28–2.49 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (int4, ~41 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (675B params, 41B active per token) + 25% KV-cache headroom ≈ 844 GB
  • 8× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~20,662 output tok/s aggregate at batch 591 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~207,439 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)~45.0%
  1. #1Claude Fable 5.293.0%
  2. #2Claude Opus 5.589.9%
  3. #22Muse Glimmer51.2%
  4. ~#23Mistral Large 3Estimated45.0%
  5. #23Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#42 of 46 (estimated)~38.6%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #41Laguna S 2.140.4%
  4. ~#42Mistral Large 3Estimated38.6%
  5. #42Nex-N2.5-Mini36.1%
No published score — estimated from related benchmarks
~#25 of 26 (estimated)~27.1
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #24GPT-6 Luna31.1
  4. ~#25Mistral Large 3Estimated27.1
  5. #25Muse Spark Open26.3
No published score — estimated from related benchmarks

Writing & chat

Vision

Hard reasoning

Agents