BREAKING
DeepSeek cuts cached input price 90%
0$/M
cache hit
0$/M
cache miss
0$/M
output
Why it matters for AI agents
1Static system prompt
2Fixed tool definitions
3RAG context reused
4Served near-free
V4-Pro vs Grok 4.5 at scale
V4-Pro63
Grok 4.5757
The caveats behind the cuts
Upside
Blended near $0.18 per million
1M-token context window
Caveats
Surge pricing at peak hours
Poor prompts erode cache gains
Will rivals cut cache prices too?
AI NEWS BLITZ
DeepSeek has slashed cached input prices by roughly 90 percent, shaking up AI agent economics.