MiniMax M2 vs MiniMax M3

Wins 0 of 5 benchmarks
$0.51 / task
Wins 2 of 5 benchmarks
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
MiniMax M213.7
MiniMax M339
Terminal-Bench 4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite
MiniMax M2—
MiniMax M32
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
MiniMax M2—
MiniMax M321.3
Arena Text · creative writing Arena Text · creative writing: Human pairwise preferences in the stated text vote pool
MiniMax M21,283
MiniMax M31,405
MMMU-Pro (vision) MMMU-Pro (vision): College-level questions that require understanding images: charts, diagrams, photos
MiniMax M2—
MiniMax M378.6

Specs

SpecMiniMax M2MiniMax M3
Input, per 1M$0.30$0.30
Output, per 1M$1.20$1.20
Cost per task?—$0.51
Speed, tok/s119157
Context205K524K
WeightsOpenOpen
ReleasedOct 2025Jun 2026
Memory (4-bit)137 GB256 GB
Licencemodified-mitminimax-community
Commercial useRestrictedRestricted
See images—✓
Read PDFs—Partial
Hear audio——
Watch video—✓
Call tools✓✓
Force a tool callPartialPartial
Return strict JSON✓—
Search the web??
Run code??
Use a computer??

What breaks when you switch

MiniMax M2 → MiniMax M3
Memory (4-bit): 137 GB → 256 GB
No strict JSON
Deployment status experimental · Self-hosted›

The documented MiniMax M3 baseline uses an experimental SGLang development image; pin its model revision and image digest and validate your workload.

MiniMax M3 → MiniMax M2
Context: 524K → 205K
No image input