Open data
Every number on the site comes from these JSON files. They are free to download, need no key or sign-up, and are rebuilt every night around 03:00 UTC; this copy was generated on November 16, 2026.
Files
| File | What it holds | Size |
|---|---|---|
| models.json | Every tracked model: maker, modalities, context window, output limit, open weights, API status, release and retirement dates. id, name, provider, modalities, contextWindow, openSource, apiAvailable, releaseDate | 202 KB |
| results.json | One row per model and benchmark: the upstream score, its 0–100 normalised value and the date of the source data. modelId, benchmarkId, rawScore, normalizedScore, sourceDate | 330 KB |
| benchmarks.json | Benchmark definitions: category, unit, direction, contamination label and source. id, name, category, higherIsBetter, scoreUnit, contaminationRisk, source | 21 KB |
| pricing.json | API list price per model in USD per million tokens, with cached-input and batch discounts and the date each price was checked. modelId, inputPerMTokens, outputPerMTokens, cachedInputPerMTokens, lastVerified, priceSource | 60 KB |
| endpoints.json | Hosted endpoints per model: price, context length, quantization and 24-hour uptime. Third-party hosts are anonymised. modelId, providerName, inputPerMTokens, outputPerMTokens, contextLength, quantization, uptime1d | 361 KB |
| scenarios.json | The scenario weightings behind the leaderboard's use-case views. id, name, description, modality, weights, cost | 14 KB |
| speed_history.json | Output speed, time to first token and API uptime: 14-day medians and daily readings per model. windowDays, asOf, models[id].tpsMedian, ttftMedianMs, successRate, daily | 120 KB |
| reliability_history.json | Daily reliability axes (output stability, format adherence, recovery, safety) for the last 30 days. generatedAt, windowDays, history[id][].date, axes | 161 KB |
| snapshots-index.json | Dates of the nightly snapshots. Each one is at /data/snapshots/YYYY-MM-DD.json with normalised scores, and prices from October 2026. dates; snapshot: date, results, pricing, models | 4 KB |
| meta.json | When the data was generated, the pipeline version and per-source freshness. generatedAt, pipelineVersion, sources, freshness | 2 KB |
| summary.json | The site's derived numbers in one small file: Analyzer Index and rank, blended price, cost per task, speed, retirement and each scenario's top 30. Used by the MCP add-on. version, generatedAt, models[].analyzerIndex, indexRank, blendedPerMTokens, costPerTaskUSD; scenarios[].ranking | 149 KB |
| bundle.json | Everything above except the history files, in one download. models, benchmarks, pricing, scenarios, results, endpoints, meta | 1.1 MB |
Using the files
Files are served with open CORS, so a web page can fetch them directly. They are cached at the edge for an hour; nightly snapshots never change once written.
const res = await fetch("https://modelanalyzer.ai/data/results.json");
const results = await res.json();
const hle = results.filter((r) => r.benchmarkId === "humanitys-last-exam");We add fields when the site needs them and try not to rename or remove existing ones. If you depend on a field, check meta.json's pipelineVersion, which changes when the shape does. For scheduled jobs, one request a day is enough: nothing changes in between.
Use it from your coding assistant
The modelanalyzer-mcp add-on lets assistants that support the Model Context Protocol (Claude Code, Cursor, VS Code and others) recommend, compare and price models and check retirement dates from this data. It runs on your machine and reads summary.json; there is no key and nothing to sign up for. The source is on GitHub (MIT).
{
"mcpServers": {
"modelanalyzer": { "command": "npx", "args": ["-y", "modelanalyzer-mcp"] }
}
}Licence and credit
Scores are gathered from public projects, and each stays under its own terms. The About page lists every source with its licence. In particular, speed and time-to-first-token figures come from Artificial Analysis and must credit them, and Arena Elo and Epoch AI scores are CC-BY 4.0.
Our own hand-maintained data (prices, speed fallbacks, hardware and image-arena seeds) is released under CC-BY 4.0. Please credit “AI Model Analyzer (modelanalyzer.ai)” and link back. Normalised scores and the Analyzer Index are computed by us; the method is on the About page.