BREAKING
Firms Slash AI Bills 90%+
Claude output price per 1M tokens
Opus
25
Sonnet 4.5
15
Haiku 4.5
5
The hybrid routing strategy
1
Flagship for complex tasks
↓
2
Auto-route by task
↓
3
Cheap models for volume
0
%
Haiku SWE-bench
0
%
Sonnet SWE-bench
0
%
GPT-4o mini MMLU
Where each model still fits
Cheaper models
good enough
●
Classification, extraction
●
Simple coding and RAG
●
High-volume processing
Flagship models
premium
●
Deep analysis
●
Advanced coding
●
Complex agent tasks
AI price war heats up
AI NEWS BLITZ
Some Anthropic and OpenAI customers are cutting AI spending by more than ninety percent.