Launching today
modelsentiment
Which AI model does Reddit actually like?
1 follower
Which AI model does Reddit actually like?
1 follower
How Reddit feels about each AI model, per version (Claude Opus 5.5, GPT-6 Sol, Gemini, DeepSeek, Qwen and more), as a 0–100 index over the last 7 days. Every number carries its 95% interval and the number of opinions behind it, so you can see when a lead is real. Each model page shows the reading day by day, what people praise and criticise, and its Arena and benchmark ranks beside it: opinion, not capability. Free, no signup, never shows post text.
Hi, I'm Henrik, a solo maker in Norway.
Every day on Reddit, someone says a model got nerfed and someone else says it is the best model ever. I wanted to know which way the room actually leans, and nobody was counting per model version, with error bars. So I built something that counts.
How it works:
- It reads the public feeds of 25 AI subreddits, one request at a time, at most one a minute.
- Code finds the model versions each post or comment names. A small language model, then DeepSeek V4.1 Flash, decides whether it holds the author's own opinion of that model and scores it.
- The scores pool into a 0–100 index per version over 7 days, with a 95% interval clustered by thread and the number of opinions behind it.
What it is not: a benchmark. It measures what people say, not what a model can do, so the Arena rank sits beside it, and the two often disagree. It never shows post or comment text, only numbers, thread titles and links.
One honest limit: the scoring is not perfect. In a blind check, 79% of counted opinions were really the author's own opinion of that model, and the ± does not include that error. The full method and its limits are on the About page.
A question for anyone who uses these models daily: does the board match your sense of the past week, and where does it look most wrong?