Back to model catalog

Compare models

Context, limits, capabilities, price and use cases side by side — up to 4 models at a time. The URL holds your selection, so you can share it.

MiniMax-M3.1-FlashMiniMax-M3Chat with these 2
AttributeMiniMax-M3.1-Flashminimax/minimax-m3.1-flashMiniMax-M3minimax/minimax-m3
Specs
Context window1M1M
Max output524K512K
Input modalitiesTextImageVideoTextImageVideo
Output modalitiesTextText
Capabilities
StreamingSupportedSupported
Tool callingSupportedSupported
Structured output (JSON)SupportedSupported
Image inputSupportedSupported
Thinking modeSupportedSupported
Price / 1M tokens
Input$ 0.15$ 0.33>= 512K$ 0.66
Output$ 0.60$ 1.32>= 512K$ 2.63
Cache read$ 0.015$ 0.066>= 512K$ 0.13
Cache write$ 0.15$ 0.33>= 512K$ 0.66
What it is good at
Best for
  • Coding agents
  • Computer use and workflow automation
  • Agents that read images and video
  • Fast agent loops
  • Multimodal agents
  • Computer use and cowork
  • Long-context research
Published benchmarks
Terminal-Bench 2.1pass_rate—
  • 66%Terminus 2(setup not fully disclosed)
Terminal-Bench 4.0pass_rate
  • 38.1%Codex 0.149.1(setup not fully disclosed)
—
Agents' Last Examscore
  • 29.6%(setup not fully disclosed)
—
AutomationBench v1.0.6score
  • 62.8%(setup not fully disclosed)
—
DeepSWE 1.1resolved_rate
  • 73.7%mini-SWE-agent 2.4.5(setup not fully disclosed)
—
NL2Reposcore
  • 61.2%Claude Code 2.1.233(setup not fully disclosed)
—
SWE-Bench Proresolved_rate—
  • 59%Claude Code(setup not fully disclosed)
SWE-Bench Verifiedresolved_rate—
  • 80.5%(setup not fully disclosed)
OSWorld 2.0success_rate
  • 61.7%(setup not fully disclosed)
—
CyberGymscore
  • 82.2%Claude Code 2.1.263(setup not fully disclosed)
—
PostTrainBench 1.1score
  • 39.3%Claude Code 2.1.251(setup not fully disclosed)
—
BabyVision (with tools)score
  • 91%Claude Code 2.1.285(setup not fully disclosed)
—
MMMU Pro (no tools)accuracy—
  • 78.1%(setup not fully disclosed)
MCP Atlas Publicscore—
  • 74.2%(setup not fully disclosed)