BREAKING
MiniMax Releases M3 With 1M Context
0
B
Total params
0
B
Activated
0
M
Max context
Staged Release in June 2026
1
API live June 1
↓
2
Open weights ~June 12
↓
3
NVIDIA free endpoint
MSA Cuts Long-Context Cost
Prefill speedup
9.7
Decode speedup
15.6
0
%
SWE-Bench Pro
0
%
Terminal-Bench 2.1
0
%
MCP Atlas
Promising, Pending Verification
AI NEWS BLITZ
China's MiniMax has launched M3, a long-context multimodal model free to try on NVIDIA.