Blog
Notes from building a scratchpad that sends one prompt to several LLMs and diffs the answers.
GPT vs Claude vs Gemini: the same 3 prompts, real answers
I ran three prompts through three models and pasted the actual output. Latency, tokens, and cost included.
Why do LLMs give different answers to the same question?
Sampling, training cutoffs, post-training, system prompts, reasoning budgets, and routing. Six reasons, no magic.
How to compare LLM outputs side by side (free, no signup)
What to look at, how word diffs help, how to pick a base model, and the mistakes I keep making.