Skip to content
Model Pareto

MiniMax-M3

MiniMaxOpen weightsminimax.io ↗
Compare with nearest rivalsCompare with…
Input
$0.30/M tok
Output
$1.20/M tok
Blended
$0.525/M tok · 3:1
Speed
118tok/s
Context
1Mtokens

Open weights (minimax-community license). Price includes a permanent 50% discount.

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
~20,773 tok/s
many requests batched
Price
$0.18–1.46 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (int4, ~73 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • FP8 weights (428B params, 23B active per token) + 40% KV-cache headroom ≈ 599 GB
  • 8× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~20,773 output tok/s aggregate at batch 747 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~162,622 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

DeepSWE v1.1Estimated
~#27 of 31 (estimated)~53.8%
  1. #1DeepSeek V4.1 Flash74.2%
  2. #2Grok 4.772.6%
  3. #26Nex-N2.5-Pro55.8%
  4. ~#27MiniMax-M3Estimated53.8%
  5. #27Qwen3.8 Max (0902)51.0%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#6 of 20 (estimated)~80.6%
  1. #1Qwen3.8-Omni-Flash92.6%
  2. #2Qwen3.8 Flash-Next91.9%
  3. #5gpt-oss-120b87.8%
  4. ~#6MiniMax-M3Estimated80.6%
  5. #6Gemma 4 31B80.0%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~#15 of 15 (estimated)~41.0
  1. #1Claude Opus 5.566.0
  2. #2Claude Fable 5.162.2
  3. #14GPT-6 Luna41.1
  4. ~#15MiniMax-M3Estimated41.0
No published score — estimated from related benchmarks

Writing & chat

MMMLUEstimated
~#2 of 5 (estimated)~89.1%
  1. #1Gemini 3.1 Pro92.6%
  2. ~#2MiniMax-M3Estimated89.1%
  3. #2Gemma 4 31B88.4%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#30 of 531427 Elo
  1. #1Claude Opus 5.51501
  2. #2Claude Fable 5.1Few votes1496
  3. #29Qwen3.8 27B1429
  4. #30MiniMax-M31427
  5. #31Gemma 4 26B A4BFew votes1423
lmarena.ai · 2026-09-24
LMArena Text (Elo)Verified
#27 of 531441 Elo
  1. #1Claude Opus 5.51509
  2. #2Claude Fable 5.11501
  3. #25Qwen3.5 397B A17B1442
  4. #27MiniMax-M31441
  5. #28Inkling1440
lmarena.ai · 2026-09-24
EQ-Bench Creative Writing v3 (Elo)Estimated
~#10 of 26 (estimated)~1863 Elo
  1. #1GPT-6 Astra2173
  2. #2Claude Fable 5.12162
  3. #9Muse Spark 1.31906
  4. ~#10MiniMax-M3Estimated1863
  5. #10GPT-5.6 Terra1855
No published score — estimated from related benchmarks

Vision

MMMUEstimated
~#2 of 3 (estimated)~71.3%
  1. #1Claude Haiku 4.573.2%
  2. ~#2MiniMax-M3Estimated71.3%
  3. #2Llama 4 Maverick69.4%
No published score — estimated from related benchmarks
CharXiv ReasoningEstimated
~#5 of 9 (estimated)~79.3%
  1. #1Gemini 3.8 Flash86.2%
  2. #2Kimi K384.8%
  3. #4Qwen3.8-Omni-Flash83.5%
  4. ~#5MiniMax-M3Estimated79.3%
  5. #5Muse Glimmer78.8%
No published score — estimated from related benchmarks
ChartographyEstimated
~#15 of 23 (estimated)~19.7%
  1. #1GPT-6 Astra71.0%
  2. #2Claude Opus 5.566.3%
  3. #14Grok 4.620.5%
  4. ~#15MiniMax-M3Estimated19.7%
  5. #15Muse Glimmer17.2%
No published score — estimated from related benchmarks

Hard reasoning

AIME (latest)Estimated
~#2 of 25 (estimated)~95.8%
  1. #1Inkling97.1%
  2. ~#2MiniMax-M3Estimated95.8%
  3. #2Muse Glimmer94.7%
No published score — estimated from related benchmarks

Agents

OSWorld-VerifiedEstimated
~#10 of 12 (estimated)~70.6%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #9Nex-N2.5-Mini71.2%
  4. ~#10MiniMax-M3Estimated70.6%
  5. #10Muse Glimmer65.9%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. #4GPT-5.547.5%
  4. ~#5MiniMax-M3Estimated<47.5%
No published score — estimated from related benchmarks
Not comparable across labs (1)

Kept for reference, not counted in rankings: these use a lab-specific task set, answer key or scoring, so scores can't be compared fairly across labs. Only published scores are shown.

OSWorld-Verified (task subset)Lab-reported
#1 of 170.1%
  1. #1MiniMax-M370.1%
minimax.io · 2026-06-01