Skip to content
Model Pareto

Mistral Small 4

Mistral AIOpen weightsmistral.ai ↗
Input
$0.15/M tok
Output
$0.60/M tok
Blended
$0.262/M tok · 3:1
Speed
170tok/s
Context
256Ktokens
Cost per task
~$0.25/taskest.

Open weights (Apache 2.0).

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
2× H200 141GB
FP8 weights
Throughput
~8,963 tok/s
many requests batched
Price
$0.088–0.73 /M tok
busy → light use
On your own machine
Mac Studio M5 Max 128GB (int4, ~130 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (119B params, 6.5B active per token) + 25% KV-cache headroom ≈ 149 GB
  • 2× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~8,963 output tok/s aggregate at batch 322 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~143,858 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)~47.2%
  1. #1Claude Fable 5.293.0%
  2. #2Claude Opus 5.589.9%
  3. #22Muse Glimmer51.2%
  4. ~#23Mistral Small 4Estimated47.2%
  5. #23Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#41 of 46 (estimated)~40.7%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #40Qwen3.8 27B42.2%
  4. ~#41Mistral Small 4Estimated40.7%
  5. #41Laguna S 2.140.4%
No published score — estimated from related benchmarks
~#25 of 26 (estimated)~27.9
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #24GPT-6 Luna31.1
  4. ~#25Mistral Small 4Estimated27.9
  5. #25Muse Spark Open26.3
No published score — estimated from related benchmarks

Writing & chat

Vision

Hard reasoning

Agents