BREAKING
Inception Launches Mercury 2
0
tok/s
Generation speed
0
x
Faster than speed-optimized
0
k
Context window
Diffusion vs Autoregressive
1
Start from noise
↓
2
Parallel refine
↓
3
Final output
0
%
IFBench
0
AIME 2025
Pricing Per Million Tokens
Input
0.25
Output
0.75
Speed Meets Reasoning
AI NEWS BLITZ
Inception Labs has released Mercury 2, a reasoning model built on diffusion.