UnifyBench
Compare AI models across published benchmarks, with weights you choose.
Our review
Checked 2026-10-02An LLM leaderboard that combines many published benchmarks instead of trusting one - and tells you to check its work.
UnifyBench scores models by consensus across multiple public benchmarks rather than reproducing a single leaderboard, with separate rankings for overall, coding, reasoning, agentic and knowledge work. Since models are often strong in one area and ordinary in another, a single number is usually the wrong question, and splitting it is the right call.
What stands out is the honesty. It labels its scores experimental, says plainly that small gaps may not be meaningful, dates the evidence it used, and explicitly tells you to shortlist here and then validate elsewhere before deciding. That is an unusual thing for a ranking site to say about itself, and it makes the rankings more useful rather than less.
It is free to browse with no account. The underlying limitation is inherent: it synthesises other people's benchmarks, so it inherits their flaws, and published benchmarks are an imperfect proxy for whether a model is good at your particular job. Use it to narrow a shortlist, then test on your own task.
Best for
- Narrowing a model shortlist before testing properly
- Comparing models on a specific kind of work rather than overall
Where it falls short
- Inherits the flaws of the benchmarks it aggregates
- Scores are explicitly experimental; small gaps are not meaningful
- No substitute for evaluating on your own task
About UnifyBench
UnifyBench ranks language models by combining results from published, sourced benchmarks into a single comparison. You can set optional weights for the capabilities that matter to your use case, compare models by performance within a budget using sourced costs, and read the methodology behind the ranking. Where a model has no published result for a benchmark, it is shown as unknown rather than guessed. The overall ranking is labelled experimental and uses a fixed reference panel. Free to use, no sign-up.
Key features
- Rankings built from sourced, published benchmarks
- Optional capability weights you set
- Performance-within-budget view using sourced costs
- Missing results shown as unknown, not estimated
- Published methodology and coverage data
Know a tool that should be here?
Submit any AI tool — yours or one you love. Free listing, no link demands, reviewed within 7 days.
Submit a tool