Founder & editor
Dylan Bristot
I built WhatLLM to make model selection less dependent on launch claims and one-number leaderboards. I write about benchmarks, model economics, open weights, coding agents, and the physical infrastructure beneath AI.
What I analyze
Which model is good for a specific job—not merely which model has the highest aggregate score.
How I work
Independent benchmark data, primary model documentation, transparent derived metrics, and explicit limitations.
What I avoid
Fixed brand verdicts, unsupported “best” claims, and comparisons that confuse a base model with an agent system.
Selected work
Analysis and methodology
The open-model gap is not one gap
A benchmark-by-benchmark analysis of the open-weight and proprietary frontier.
Which coding LLM benchmark should you trust?
SWE-bench, Terminal-Bench, LiveCodeBench, and SciCode mapped to the jobs they measure.
The cheapest coding model is not always the cheapest
A transparent framework for model cost, retries, review, and accepted results.
Cost per task is the new agentic AI benchmark
Why agent economics need a better unit than token price.
The world is built out of a few narrow places
A long-form analysis of the physical constraints beneath the AI buildout.
WhatLLM methodology
How model quality, coding, agents, speed, price, and provider data are handled.