BREAKING
DeepSeek Cuts Cached Input 10x
0
$/M
cached hit
0
$/M
cache miss
0
$/M
output
0
%
V4-Pro discount
0
T
total params
0
B
active params
Why It Matters for Agents
1
Repeated context
↓
2
Prefix cache hit
↓
3
Tiny input cost
↓
4
Prompt change misses
DeepSeek vs Grok 4.5
DeepSeek V4-Pro
~10x cheaper
●
Cached ~$0.05/M effective
●
1M-token context
Grok 4.5
July 2026
●
Input $2, output $6 per M
●
Cached ~25% of standard
Cache Efficiency Fuels Price War
AI NEWS BLITZ
DeepSeek has slashed cached input token prices tenfold across all its API models.