# Benchize > Source: Benchize (https://benchize.com) — https://benchize.com/. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/. Public benchmark reports, structured with original sources and evidence. Canonical URL: https://benchize.com/ Last updated: 1970-01-01T00:00:00.000Z Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations. ## How to use this site Benchize collects public LLM benchmark reports. Start with the model directory to find evaluations for a model, or the benchmark directory to compare reported model results for a task. Benchize preserves original evidence and does not run model inference or calculate a new composite score. Public pages are currently English; the homepage redirects to /benchmarks. Read the [Markdown & JSON API guide](https://benchize.com/docs.md) for a step-by-step introduction, URL builder, code examples, pagination, caching and citation guidance. 1. Discover identifiers in [Models](https://benchize.com/models.md), [Benchmarks](https://benchize.com/benchmarks.md), or the complete section indexes linked below. The selected entries below are the first 10 from each unfiltered directory, in the same order. Model order follows release date; benchmark order prioritizes published model coverage and recent models. Complete indexes retain that order across pages. 2. Read one model at /models/.md or /models/.json; read a benchmark at /benchmarks/.md or /benchmarks/.json. Detail documents do not have HTML pages. Use the listed identifiers verbatim and URL-encode them. Public configuration suffixes use a single hyphen, such as -thinking, -high or -high-budget-32768; unspecified configurations have no suffix. Copy existing IDs instead of guessing suffixes from names. 3. Markdown is for reading and citation; JSON is for structured processing and preserves metrics, configuration, statistics, provenance, version and timestamps. Public HTML directories also support Accept: text/markdown; append .md for an explicit Markdown URL. Browser directory clicks open dialogs without changing the URL. Capability categories use /benchmarks/category/, including coding and instruction-following. ## Query and pagination - Find models with `GET /api/v1/models?search=&developer=&limit=100`; developer is optional. The HTML/Markdown directories accept the same search and developer filters. - Find benchmarks with `GET /api/v1/benchmarks?search=&category=&limit=100`; category is optional. Use the Categories links below for canonical category slugs. - Public list APIs accept limit=1..100. When a JSON response has a non-null nextCursor, URL-encode it and request the same path with cursor=, retaining the filters. Stop when nextCursor is null. Detail JSON documents return up to 100 reports and also provide a Link header with rel="next"; Markdown details link to their next page. Do not assume the first page contains every report. - /llms/models.md and /llms/benchmarks.md list every eligible detail document. /llms-full.txt links to bounded five-page content shards; follow each shard's source and continuation links when more reports are needed. Unknown or unpublished documents return 404; malformed JSON-document parameters return 422. Public reads need no API key; /internal/ is private. ## Interpret results and freshness Compare reports only when benchmark, score publisher, sourceVersion, metric key, unit, scope and direction are compatible. A percent and a score_fraction are different raw units. Keep original configurations, uncertainty labels and individual sourceUrl citations; do not treat an unreported configuration as disabled thinking or a missing measurement as zero. Report dates distinguish evaluatedAt from observedAt. Cite Benchize and the original source evidence together. Completed source collections refresh the catalog. Last updated is the latest completed data revision; individual report timestamps describe their own source observations or evaluations. There is no guaranteed update cadence for every source. ## Docs and API - [Markdown & JSON API](https://benchize.com/docs.md): Read LLM benchmark reports with a URL. Use Markdown with AI assistants, or JSON in your own tools, with original sources and no API key. ## Categories ## Leaderboards/benchmarks - [Complete benchmarks index](https://benchize.com/llms/benchmarks.md): every eligible detail document in directory order. ## Models - [Complete models index](https://benchize.com/llms/models.md): every eligible detail document in directory order. - [Main index](https://benchize.com/llms.txt): usage instructions and directory-ranked selections. - [Full content](https://benchize.com/llms-full.txt): complete source-linked page content. --- Cite as: Benchize — Benchize, https://benchize.com/ (retrieved 2026-10-21)