state of the art
In plain wordsThe best performance or most advanced technique currently known in a field. In AI papers, a state-of-the-art result is one that beats every earlier published result on a test.
- 1
AI research
The best published result on a given benchmarkbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time. or task at a given time, or the method that achieves it; abbreviated SOTA. Machine-learning papers conventionally report gains over the prior best, and leaderboardsleaderboard, n.: A public ranking table showing which AI systems score best on a test, updated as new results arrive — like a sports league table for models. track who holds it. Critics argue that SOTA-chasing rewards small benchmark gains, often within noise or obtained with more compute or tuning than the baselinesbaseline, n.: A simple reference method that a new model must beat to show it is worth anything, such as always guessing the most common answer or reusing yesterday's value. received, over understanding, and that benchmarks saturate or leak into training data (benchmark saturationbenchmark saturation, n.: When the best AI systems score so close to the maximum on a test that it can no longer tell them apart or show further progress., data contaminationdata contamination, n.: When the test questions used to grade an AI system were already in its training data, so a high score may reflect memorized answers rather than real ability.). A claim to hold the state of the art should therefore name the benchmark, the comparison set and the date.
The paper claims state of the art on three benchmarks, but the baselines were run with default settings.
- 2
Engineering and lawlegal
The most advanced knowledge and practice available in a technical field at a given time. In patent law it denotes the prior art against which novelty and inventive step are judged; under the European Patent Convention it comprises everything disclosed to the public, by any means, before an application's filing date (Art. 54(2)). Regulations also use it as a standard of care, as when the EU AI ActEU AI Act, prop. n.: The European Union's law on artificial intelligence, the first broad AI law anywhere: it bans some uses, sets strict checks for risky ones and puts duties on the makers of the biggest AI models. requires that measures take into account the 'generally acknowledged state of the art'.
Origin
Recorded in English engineering writing in 1910, the Oxford English Dictionary's earliest citation coming from Henry Harrison Suplee's book on gas turbines; the hyphenated adjective developed later. The acronym SOTA became pervasive in machine learning in the 2010s with benchmark-driven deep-learning research.
Usage note
The hyphenated adjective is common in marketing, where it means little, and this dictionary avoids it as praise. A SOTA result is relative to a benchmark and a date and is often superseded within months.
Related
- benchmarkbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time.
- leaderboardleaderboard, n.: A public ranking table showing which AI systems score best on a test, updated as new results arrive — like a sports league table for models.
- benchmark saturationbenchmark saturation, n.: When the best AI systems score so close to the maximum on a test that it can no longer tell them apart or show further progress.
- baselinebaseline, n.: A simple reference method that a new model must beat to show it is worth anything, such as always guessing the most common answer or reusing yesterday's value.
- data contaminationdata contamination, n.: When the test questions used to grade an AI system were already in its training data, so a high score may reflect memorized answers rather than real ability.
- benchmaxxingbenchmaxxing, n.: Slang for tuning an AI model to score as high as possible on well-known tests, even when this does not make it genuinely better at real-world tasks.
- frontier modelfrontier model, n.: One of the most capable AI models in existence at a given time, built by a handful of leading labs at great cost, and the main target of AI safety rules.
- baselinebaseline, n.: A simple reference method that a new model must beat to show it is worth anything, such as always guessing the most common answer or reusing yesterday's value.
Defined in law or standards
European Patent Convention, Art. 54(2) (patent-law sense)
Mentioned in
baselinebaseline, n.: A simple reference method that a new model must beat to show it is worth anything, such as always guessing the most common answer or reusing yesterday's value., Papers with CodePapers with Code, prop. n.: A free website, run by Meta from 2019 until it closed in 2025, that linked AI research papers to their computer code and kept leaderboards of the best results.