More statistics

Deeper breakdowns of how tokens are consumed across models, the team's experience profile, and the features being shipped.

Model mix

5 models in production

Team profile

15 developers · 4 levels

Revenue impact

Tracked over 6 months

Tokens vs AI models

Where the team's token budget actually goes, by underlying model

Claude Sonnet 4.5

4.3B$13,042

GPT-5

2.8B$6,987

Gemini 2.5 Pro

1.7B$2,981

Claude Haiku

1.0B$259

GPT-5 Mini

517.5M$207

Premium reasoning models (Sonnet, GPT-5) drive ~70% of spend while only powering ~45% of completions. A routing strategy could shift simple tasks to Haiku / Mini.

Tokens by developer role

Which type of developer relies the most on AI assistance?

DevOps· 7 devs

5.5B

52.9% of team spend · avg 781.5M per dev

Fullstack· 3 devs

2.0B

19.2% of team spend · avg 662.7M per dev

Frontend· 2 devs

1.7B

16.7% of team spend · avg 865.6M per dev

Backend· 3 devs

1.2B

11.2% of team spend · avg 387.0M per dev

Frontend and Fullstack roles tend to consume the most tokens — UI iteration and cross-stack glue work both benefit heavily from AI assistance, while DevOps work is more script-driven and needs fewer completions.

Tokens vs seniority

Are senior devs more efficient with AI — or do juniors lean on it more?

Average monthly tokens by seniority bucket

Mid-level developers (3-5y) burn the most tokens on average — they're productive enough to ship a lot, but not yet senior enough to know when AI assistance is overkill.

Tokens spent vs total revenue

Every $1 spent on tokens this half-year generated ${ratio} of company revenue.

Token spend (6 mo)

$47,200

Revenue (6 mo)

$1,496k

Revenue per $1 of tokens

$32

Revenue grows faster than token spend — clear positive ROI on AI investment.