← All models

MiniMax: MiniMax M3

MiniMax

open_source_balanced

Capabilities

VisionToolsStructured OutputExtended ThinkingLong ContextCode

Strengths

  • Exceptional 1M-token context window for long-horizon agentic tasks
  • Strong multimodal capability (text, image, video inputs)
  • Open-weight model enabling full data privacy and self-hosting
  • Excellent instruction-following and conversation performance
  • Cost-effective pricing ($0.3/$1.2 per M tokens)

Weaknesses

  • Very slow inference (0.06 speed score, ~66 tok/s) limits real-time applications
  • Video understanding capabilities not independently validated
  • Training data composition not fully disclosed
  • Smaller provider means less ecosystem tooling than Anthropic/OpenAI

Pricing

Input / 1M tokens

$0.3

Output / 1M tokens

$1.2

Context window

1,048.576k

Transparency

Open weights

0.0 / 10

Open training data

0.0 / 10

Open methodology

6.5 / 10

Licence openness

3.0 / 10

Provider disclosure

6.0 / 10

FMTI company score

N/A

Composite:4.0 / 10

MiniMax ships open-weight models, enabling self-hosting and data privacy. Training methodology partially disclosed but data provenance not fully transparent. Strong licence openness reflects permissive open-model licensing.

Sustainability

Inference energy

N/A

Training footprint

N/A

Provider infrastructure

5.0 / 10

Composite:5.5 / 10

Limited public data on MiniMax infrastructure sustainability. 1M context window and multimodal capability suggest higher inference costs, but open-weight model enables efficient self-hosted inference optimisation.

Task Fitness

question answering
82%
code
82%
mathematics
91%
business communication
90%
vision
78%
analysis
84%
extraction
90%
long-form generation
86%
research
87%
reasoning
82%
summarisation
85%
translation
80%
conversation
92%