Don't worry, it's not broken!
If you clicked a link like /v1/leaderboard.json and saw a wall of raw brackets and text, that's completely normal! That format is called JSON (JavaScript Object Notation), and it's how software applications, Python scripts, and AI agents "talk" to each other without human interface lag.
The Scoreboard Analogy: Think of this endpoint like an automated sports scoreboard wire. Instead of a human opening a browser to look at a webpage, your code can pull down the exact real-time rankings, ELO scores, fuel prices, and porosity ratings across all 10,858 AI engines with a single command.
Run this in Python to query the leaderboard directly:
Fetch in modern JavaScript / Node.js:
Command line terminal query with jq:
Import live telemetry directly into your spreadsheet:
Live Endpoint Inspector: /v1/leaderboard.json
Below is a live, formatted preview of the actual payload served to applications.
Field-by-Field Decoder Guide
What every JSON key in the leaderboard payload means:
| JSON Key | Data Type | Plain-English Meaning | Example |
|---|---|---|---|
| rank | Integer | Current overall position on the Pit Wall grid based on verified telemetry. | 1 |
| dossier_id | String | Canonical path handle linking maker, architectural family, and model SKU. | "deepseek/v4/deepseek-v4-flash" |
| name | String | Clean human-readable title of the AI engine. | "DeepSeek V4.1 Flash" |
| context_length | Integer | Maximum token memory capacity (displacement) the engine can process in one turn. | 128000 |
| prompt_price_per_m | Float ($) | Cost in USD to send 1 Million prompt tokens to the model. | 0.14 |
| empirical_score | Integer / Null | COSMOS 100-Point Dyno Score evaluated across 6 ground-truth harness dimensions. | 67 |
| coding_elo | Integer / Null | Arena Elo rating specifically measured on verified coding benchmarks. | 1330 |
| swe_bench | Float (%) | Percentage of real-world GitHub issues solved and verified under automated testing. | 38.8 |