| name | artificial-analysis-leaderboard |
| description | Fetch live LLM model rankings and Artificial Analysis composite indices — Intelligence Index, Coding Index, and Agentic Index — plus speed and pricing from https://artificialanalysis.ai/leaderboards/models. Use when the user asks for model benchmarks, "best models", model comparison, leaderboard/ranking, "가장 성능 좋은 모델", "모델 순위", "코딩 잘하는 모델", "에이전트/코딩 점수", Intelligence/Coding/Agentic Index, model speed/latency/pricing, or which model to pick for a task. Returns current (live) data, not cached. |
Invoke as a skill (e.g. /skill:artificial-analysis-leaderboard) or just ask about model
rankings in natural language (e.g. "요즘 가장 똑똑한 모델 뭐야?"). The script runs under
Node.js with zero dependencies — it works with any agent that loads SKILL.md.
Artificial Analysis Leaderboard (Live)
Fetches live model rankings and metrics from
https://artificialanalysis.ai/leaderboards/models and prints them as a
human-readable table or JSON.
When to use
- "요즘 가장 똑똑한 모델 뭐야?" / "best LLM right now"
- "모델 순위", "leaderboard", "model ranking"
- Model comparison by intelligence, coding, agentic, speed, price, or latency
- "코딩 잘하는 모델", "에이전트 점수", "best coding/agent model"
- "가장 빠른 모델", "가장 싼 모델", "best speed/price model"
- Filtering by creator (Anthropic, OpenAI, Google, DeepSeek, …) or model name
How it works
The page is server-side rendered (Next.js) and embeds the full per-model data payload
in the HTML — no API key, no JS rendering, no scraping of dynamic content. The skill
script (aa-leaderboard.js, Node.js stdlib only — no dependencies) fetches the page and
extracts the embedded model array (~225 models, refreshed continuously), including the
Artificial Analysis composite indices (Intelligence / Coding / Agentic).
Data fields
Each model row contains:
| Field | Meaning |
|---|
model | Model name (e.g. GPT-5.5 (xhigh)) |
creator | Vendor (Anthropic, OpenAI, Google, …) |
intelligence | Intelligence Index v4.1 — main quality score (higher = better). A trailing * means estimated. |
coding | Coding Index — SWE-Bench Verified / LiveCodeBench blend (higher = better at coding) |
agentic | Agentic Index — τ²-Bench / GDPval-AA / Terminal-Bench blend (higher = better at tool use / agents) |
blendedUSD | Blended price per 1M tokens (lower = cheaper) |
speedTokPerSec | Median output throughput, tokens/sec (higher = faster) |
latencyFirstChunk | Time to first chunk, seconds (lower = snappier) |
totalResponse | End-to-end response time, seconds |
-- means that model has no data for that metric (e.g. newly released).
Usage
The script is at ./aa-leaderboard.js (relative to this skill dir). Run it with node.
node aa-leaderboard.js --top 15
node aa-leaderboard.js --sort coding --top 10
node aa-leaderboard.js --sort agentic --top 10
node aa-leaderboard.js --sort speed --top 10
node aa-leaderboard.js --sort price --top 10
node aa-leaderboard.js --sort latency --top 10
node aa-leaderboard.js --creator anthropic --top 10
node aa-leaderboard.js --creator deepseek
node aa-leaderboard.js --model 'mini' --top 10
node aa-leaderboard.js --json --top 20
node aa-leaderboard.js --json --creator google
Flags
| Flag | Description |
|---|
--top N | Limit to top N rows (after sort + filter) |
--sort FIELD | intelligence (default) · coding · agentic · speed · price · latency · name |
--creator X | Filter: creator contains X (case-insensitive) |
--model X | Filter: model name contains X (case-insensitive) |
--table | Human-readable aligned table (default) |
--json | Emit JSON array (each row = object with the fields above) |
Sort direction: intelligence/coding/agentic/speed sort descending (best first);
price/latency sort ascending (cheapest/snappiest first); name sorts A→Z.
Rows with -- in the sort field are pushed to the bottom.
Output tips
- When the user asks a natural-language question, run the relevant query and
summarize the result in prose — don't dump the raw table unless asked.
- The site tracks several leaderboards; this skill covers the models leaderboard
(Intelligence / Coding / Agentic indices, speed, price). For image generation,
speech, video, etc., point the user to the site directly.
Reliability / notes
- Fetch is a plain HTTPS GET with a browser User-Agent. The parser extracts the
embedded model data payload (see
parse() / field() in the script); if the site
changes that payload shape, the parser will need updating.
- Data is live at fetch time; Artificial Analysis updates continuously.
- No API key required — public leaderboard.
