BREAKING
MiniMax M3 Joins Agent Arena
0
M
max context tokens
0
K
guaranteed min
0
x
faster prefill
0
x
faster decode
Vendor-Reported Benchmarks
SWE-Bench Pro
59
Terminal-Bench 2.1
66
BrowseComp
83.5
Strengths vs Concerns
Praised
●
Handles long complex tasks
●
E-commerce site builds
●
Cost-efficient long context
Concerns
●
Unpredictable and verbose
●
Prone to hallucination
●
May stop midway
Leaderboard Score Coming Soon
AI NEWS BLITZ
China's open-weight model MiniMax M3 has been added to the Agent Arena evaluation platform.