Tech Executive & AI Advisor
CTO / CIO / Consultant / Applied AI Practitioner
Explore Advisory Services →Bars: repositories with at least one commit that week — 108 at the busiest, out of 152 I build and run. 68% of commits paired with a coding agent. · commits
Bars are input tokens per week (incl. cache-creation/cache-read), left axis; the line is output tokens per week, right axis, independently scaled — not dollar cost. · output tokens
- Opus 5 +24.3pt 48.1%
- Sonnet 5 -24.4pt 44.7%
- Haiku 4.5 -0.1pt 5.9%
- Opus 4.8 -0.1pt 0.8%
- Opus 4.7 -0.1pt 0.3%
- Fable 5 0pt 0.1%
- claude-fable-5-1 0pt 0.1%
LLM Deliberation (multillm council) · 103,932 tokens, 08-31 – 09-21
- GPT-4.1 Mini 19,751 19%
- Gemini 3.1 Flash Lite 19,084 18.4%
- DeepSeek V3 17,524 16.9%
- Gemini 3 Flash Preview (synth) 17,440 16.8%
- Llama 3.3 70B 15,960 15.4%
- Mistral Small 3.1 14,173 13.6%
Long tail, tracked separately from the Claude mix above — % is a share of council/deliberation usage only, not the fleet total.