ITG NowHome › AI glossary › Benchmark
What is an AI benchmark?
A benchmark is a standard test that lets you compare how well different AI models do the same tasks.
- Category Benchmarks
- Also called benchmarks, AI benchmark
- Updated 20 Oct 2026
Benchmark, explained simply
A benchmark is a fixed set of questions or tasks. Every AI model takes the same test, so the scores can be compared.
The tests cover things like math, coding, reading and facts. Each one gives a score.
This matters because it helps you judge claims. Companies say their model is best, and benchmarks give a common yardstick.
For example, think of a school exam for AI. A high score tells you something, but one exam doesn't show everything. Models can also be trained to do well on a known test.
On ITG Now, the AI models page shows each model's score on several benchmarks, with a link to each source.
Related terms
LeaderboardBenchmark contaminationAI modelLLMSWE-benchGPQA Diamond
Where you'll see it on ITG Now
- GuideMoli: A Free, Light Headless Browser for AI Agents
- GuideHindsight: Free AI Agent Memory That Learns (2026 Guide)
- GuideOpenHuman: Open Source AI Agent App That Runs Lean (2026)
- GuideProject NOMAD: Free Offline Knowledge Server (2026 Guide)
- GuideStrata: Run Big Local AI on a Gaming PC, Free
- AI newsDeepSeek says free model update tops a coding test
Questions people ask about Benchmark
Is a higher benchmark score better?
Usually yes, but check the test. Some scores, like cost or delay, are better when lower.
Can I trust AI benchmarks?
They are useful but not perfect. Models may be tuned for them, so look at more than one.
Benchmark vs leaderboard: what is the difference?
A benchmark is the test. A leaderboard is a list that ranks models by their scores.