Which AI model is actually good at what?
Honest, independent test results — explained in plain words. Our own measurements on real tasks, plus the results of 52 public leaderboards collected in one place. No vendor marketing.
New to this? Start with the glossary — 2 minutes, no jargon.
What do you need a model for?
Who's winning right now
Most 1st places across public leaderboards. Full rankings →
- #1Claude Fable 5.211 × 1st · 16 × top 3
- #2Gemini 4 Carbon6 × 1st · 11 × top 3
- #3Claude Opus 5.54 × 1st · 9 × top 3
- #4GPT-6 Astra3 × 1st · 11 × top 3
- #5Claude Opus 53 × 1st · 5 × top 3
Our own tests
Measured by us, with dates and sample sizes. How we test, and the caveats.
Speech
- ASR: Polish speech recognitionHow well do speech-to-text models write down spoken Polish — including Polish mixed with English, noisy rooms and live speech?8 tables
- Diarization: who is speakingCan a model tell who is speaking when, in a recording with several people?2 tables
- Omni models: video and audio inputModels that claim to accept video and sound: do they really hear the audio and see the frames?2 tables
Vision & OCR
- OCR: handwritten notes, boards and textbook pagesWhich AI models can read handwritten notes, whiteboards and textbook pages from a photo — including Polish letters, formulas and small numbers?7 tables
- OCR: exam-sheet photos with handwritten answersCan models read photos of exam sheets: circled answers, handwritten words, and tell an empty page from a filled one?4 tables
Language models
- Language models: classification, summaries, chat with toolsHow good are language models at everyday text jobs: sorting a text into a category, summarizing without losing facts, chatting while using tools?6 tables
- Language models: short answers, grading text, time and costShort answers, grading a written answer, and what each call costs and how long it takes.1 table
- Reasoning level (reasoning_effort): what models actually doMany models have a 'think harder' setting. Does it actually change anything, and what does it cost?3 tables
Embeddings
Platforms
Security (macOS malware scanning)
- Malware detection: AMTSO, EICAR containers, Objective-See hashes, Mac malware familiesDoes a Mac malware scanner catch the standard harmless test files and known Mac malware families?7 tables
- Malware detection: test samples vs held-out clean filesA fair test: how much malware does the scanner catch, and how often does it wrongly flag normal files it was never tuned on?5 tables
- Malware scanning: static-heuristic false alarms before and after fixesHow often the scanner wrongly flagged normal Mac files, before and after fixes.9 tables
- ML classifiers on TLSH fingerprints of Mach-O filesCan a machine-learning model spot malicious Mac programs from a file 'fingerprint' alone?5 tables
- Scanner performance: database loading, re-scans, TLSH in SQL, Aho-CorasickSpeed measurements of a malware scanner: loading its databases, re-scanning, and two search libraries.2 tables
Public leaderboards
52 from labs and universities · 47 with results copied here.
Want to be a sponsor?
Help keep these benchmarks independent and free — or tell us what you'd like tested next. Sponsors are credited and never influence results.
Using an AI assistant? Give it /bots
Everything here is also available as plain Markdown, CSV and JSON — so your assistant can use real measurements instead of inventing its own test. Point it at benchmarks.site/llms.txt.