About ToneBench
Who built it, why, and what you can do with it.
I'm Louis-François Bouchard, co-founder and CTO of Towards AI, and I host the "What's AI" YouTube channel. I built ToneBench because I wanted to know which models can actually write a script the way I do: the whole thing, with the hook, the structure, the anti-hype voice and the technical depth, good enough to air without edits. Generic leaderboards can't answer that, so the benchmark is built on the scripts behind 10 of my own What's AI videos, and my finished version of each one is the human reference every model is measured against.
What you can use it for
If you came here with a practical question, here's where each answer lives:
- Which model should write in my voice? The leaderboard ranks every configuration on the same scripts, and each one has a scorecard that shows where it's strong, where it slips, and how consistent it is from run to run.
- Is thinking harder worth the money? The thinking-level page puts writing score next to cost and time for every reasoning-effort setting. For some models the top setting is a real jump; for others you pay several times more for almost nothing.
- Model A or model B? The head-to-head pages put two configurations side by side, with a written verdict for each pair.
- How much should I trust this? The methodology page covers the judges, the intervals, the checks, and every limit we know about. If you run evals on your own product, it's also a decent template to borrow from.
Hear the voice the models are chasing
"What's AI" on YouTube Practical, no-hype AI engineering videos on LLMs, RAG, agents, evals, and shipping real AI products. It's free, and it's where the scripts in this benchmark come from, so if you want to know what the models are being measured against, this is it. Subscribe and watch →Who is Towards AI?
We started as a Medium publication, grew into one of the bigger AI learning communities with a newsletter to match, and wrote the book Building LLMs for Production. We're engineers and researchers, and mostly we care about AI that actually ships. More about us →
I build and maintain the benchmark, with the Towards AI editorial team, who reviewed every reference script before it became the bar. You can find more of my work on my website, and how the scores are produced is on the methodology page.
Work with us
Learn with us
Hands-on courses for becoming an AI engineer who actually builds and ships, from Master AI Engineering to Agent Engineering and more. You finish with a deployed product and a certified portfolio you can show.
For your company
We train engineering teams to build with AI, and when you'd rather have us build it with you, we do that too. The case studies show what that looked like for other companies.
Frequently asked questions
What is Towards AI?
The company I co-founded, and the team behind this benchmark. We teach AI engineering, publish, and build AI with companies; the sections above have the details, and the rest is on towardsai.net. ToneBench is our writing benchmark.
Who runs the What's AI YouTube channel?
I do. I'm Louis-François Bouchard, co-founder and CTO of Towards AI. The channel is practical, no-hype AI engineering, and it's where this benchmark's reference scripts come from. Subscribe here if that's your kind of thing.
How can I learn AI engineering with Towards AI?
Start at the Academy. The Learn with us card above covers what the courses give you, plus the book.
Can Towards AI help my company adopt AI?
Yes. That's the Work with us section above: team training, custom builds, and the case studies.
How is ToneBench scored?
Every current-ranked model writes the same 10 real scripts, five times each. Every draft is scored blind against my finished version on nine dimensions: seven by a panel of three LLM judges from three model families, one (length) in code, and one (slop) half in code, half by the judges. The full details, including how far to trust the ranking, are on the methodology page, and the ranking itself is on the leaderboard.