tokens
The receipt for how I actually build with AI.
Every model I run, every tool it runs in, and the one question everyone keeps asking, answered in real numbers.
Which model is Jorge using?
Best reasoning-to-speed ratio for shipping real product. It writes code I actually merge.
The models I reach for, ranked.
Share of total usage across every tool. Different jobs, different models, I'm not loyal, I'm efficient.
- 1Claude Opus 4.8AnthropicDaily driver46%
The one I ship with. Plans, writes, and reviews entire features end-to-end. · 48B tokens
- 2Claude Sonnet 4.6AnthropicWorkhorse21%
Fast, cheap, and good enough for the 80% of edits that don't need Opus. · 22B tokens
- 3GPT-5OpenAI14%
Second opinion on tricky bugs and a different voice for brainstorming. · 14B tokens
- 4Gemini 3 ProGoogle9%
Huge-context reads, dumping a whole repo or PDF and asking questions. · 9B tokens
- 5Claude Haiku 4.5AnthropicSpeed runs7%
Inline autocomplete and quick one-liners where latency beats depth. · 7B tokens
- 6Local (Llama / Qwen)Open weights3%
Offline tinkering and anything I'd rather not send to a cloud. · 3B tokens
Five platforms, one workflow.
Pulled from each tool's usage dashboard and year-in-review. Numbers refresh as I do.
Where most of the building happens, an agent that lives in my terminal and ships whole features.
My editor of record, tab-completions and inline agents while I'm hands-on in the code.
Multi-agent runs for bigger refactors and research sweeps that one context can't hold.
Brainstorming, naming things, and quick research outside the editor. The rubber duck.
Cited, up-to-date answers when I need a source, not a hallucination.
Opinions, formed at 100B tokens.
Last synced · June 2026
