Per-page Markdown
/bots/methodology.md— Methodology & caveats (read before citing)/bots/asr-polish.md— ASR: Polish speech recognition/bots/diarization.md— Diarization: who is speaking/bots/omni-video-audio.md— Omni models: video and audio input/bots/ocr-handwriting-boards-textbooks.md— OCR: handwritten notes, boards and textbook pages/bots/ocr-exam-sheets.md— OCR: exam-sheet photos with handwritten answers/bots/llm-classification-summaries-chat.md— Language models: classification, summaries, chat with tools/bots/llm-short-answers-grading-cost.md— Language models: short answers, grading text, time and cost/bots/reasoning-effort.md— Reasoning level (reasoning_effort): what models actually do/bots/embeddings-polish.md— Embeddings: retrieval accuracy in Polish/bots/platforms-availability-latency-limits.md— Platforms: availability, latency and limits/bots/platforms-errors-billing-prices.md— Platforms: errors, reasoning billing and prices/bots/malware-detection-safe.md— Malware detection: AMTSO, EICAR containers, Objective-See hashes, Mac malware families/bots/malware-detection-held-out.md— Malware detection: test samples vs held-out clean files/bots/malware-static-false-positives.md— Malware scanning: static-heuristic false alarms before and after fixes/bots/ml-tlsh-macho-classifiers.md— ML classifiers on TLSH fingerprints of Mach-O files/bots/scanner-performance.md— Scanner performance: database loading, re-scans, TLSH in SQL, Aho-Corasick/bots/internet.md— Benchmarks from the internet
CSV
One CSV per result table: /data/.
How to use this data
- Every result is for a specific date, platform and sample size. Quote them together.
- Many samples are small and synthetic; differences of 1–3 percentage points are often noise. See methodology and caveats.
- Every HTML page links its Markdown twin via
<link rel="alternate" type="text/markdown">. - Cite as benchmarks.site with the page URL and measurement date.