About ChatJudge
ChatJudge helps you choose which AI to try for a task, and shows the evidence. We compare AI models on 21 public benchmark leaderboards, grouped by task (Best AI by task), by field and head to head (Compare models). Every page names the model or tied models with the top reported score, the original source, the leaderboard’s own date and the date we last checked it. See every leaderboard.
Original sources
We don’t run these tests. Every score is copied as published by the organisation that ran the benchmark: Legal Benchmarks, Vals AI, Artificial Analysis, Datacurve, Scale AI and Center for AI Safety and LMArena. Each page links to the original leaderboard. Model names appear as each source lists them, and where a source publishes a cost per task, we show it as the API cost per benchmark task, not a subscription price.
How we update the data
- A scheduled job re-reads the sources each morning, Singapore time. The most recent reading behind this site was on 17 Nov 2026.
- Vals AI, Artificial Analysis and Datacurve (DeepSWE) leaderboards are fetched and parsed by a script. Legal Benchmarks, Scale AI’s Humanity’s Last Exam and LMArena draw their tables in the browser, so their rows are copied from the source page by hand and checked against it on each refresh.
- Scores are never typed in from memory or estimated. If a source changes its layout, the script stops instead of publishing partial data, and the previous scores stay up.
- Leaderboard dated is the date the source gives its leaderboard. Last checked is when we last read the source for the version of the page you’re seeing. Some sources no longer run their benchmark on new models; we label those leaderboards as no longer updated or retired.
The prompt library
Each task guide includes a prompt you can copy and try with the leading models. The prompt library holds 427 prompts. Most were shared publicly on X or Threads; those are credited to their creator with a link to the original post. Others were written by us, or adapted from openly licensed projects and credited on our credits page.
Corrections
If a score, model name, date or source on ChatJudge is wrong, email hello@advark.ai with the page address, what’s wrong, and a link to the original source that shows the right value. We check it against the source, fix the page and update its date. Creators who want a prompt credited differently or removed can use the same address.
How the site is funded
ChatJudge is free to use, supported by sponsors and affiliate links. Sponsored and affiliate links are labelled, are never part of rankings or benchmark summaries, and never change a score.
For teams
Share your best prompts with your team: a private, searchable library for your startup. Set up a team library.
