SpendFirewall benchmarks
Token cost benchmarks by provider
Median input token cost: Claude Opus $15/M, GPT-4 $10/M, Gemini Pro $3.50/M, Claude Sonnet $3/M, GPT-4-mini $0.15/M. Output is 3-5x input across providers. SpendFirewall normalizes enforcement across all of them.
The data
Median input token cost: Claude Opus $15/M, GPT-4 $10/M, Gemini Pro $3.50/M, Claude Sonnet $3/M, GPT-4-mini $0.15/M. Output is 3-5x input across providers. SpendFirewall normalizes enforcement across all of them. These benchmarks come from anonymized SpendFirewall customer data and public research.
Why these benchmarks matter
If you don't know where you stand, you can't tell whether your agent spend is normal or pathological. These benchmarks give you the comparison.
How to use these benchmarks
- Measure your current spend per task, per agent, per day
- Compare to the benchmarks above
- If you are above the 75th percentile, install SpendFirewall to enforce ceilings
Frequently asked questions
Where does this data come from?
Anonymized SpendFirewall customer data plus public research from Anthropic, OpenAI, and independent surveys.
How often is it updated?
Quarterly, as the underlying provider pricing changes.