BREAKING
MiniMax-M3 Open Weights on Hugging Face
0
B
total params
0
B
active min
0
M
token context
0
x
faster prefill
0
x
faster decode
0
T
pretrain tokens
Vendor-Reported Benchmarks
SWE-Bench Pro
59
Terminal-Bench 2.1
66
SWE-fficiency
34.8
KernelBench Hard
28.8
Promise vs Caution
Strengths
●
Better agentic coding and planning
●
1M-token long context
●
Free NVIDIA endpoint for prototyping
Concerns
●
Slower inference in some setups
●
Possible SWE-Bench contamination
●
Commercial-use license ambiguity
Open Frontier Agentic AI
AI NEWS BLITZ
MiniMax has released open weights for M3, and NVIDIA added a free endpoint.