artificial-analysis-leaderboards¶

Daily snapshots of Artificial Analysis AI model leaderboards (LLMs, T2I, T2V, I2V, TTS, Image Editing)

项目信息
仓库	https://github.com/Maicarons/artificial-analysis-leaderboards
在线体验	https://artificialanalysis.ai
主要语言	Python
Stars	0
类型	原创项目
许可证	—
最近更新	2026-09-25

Artificial Analysis Leaderboards¶

Daily snapshots of Artificial Analysis leaderboard data collected from public web surfaces.

Data¶

Endpoint	Description	Data
`llms`	LLM leaderboard	Intelligence, coding, agentic, pricing, speed, context window, modality support
`text-to-image`	Text-to-image arena rankings	ELO, rank, CI95, appearances, win rate, per-tag breakdowns
`image-editing`	Image editing arena rankings	ELO, rank, CI95, appearances, win rate
`text-to-speech`	Text-to-speech arena rankings	ELO, rank, CI95, appearances, win rate, category/accent breakdowns
`text-to-video`	Text-to-video arena rankings	ELO, rank, CI95, appearances, win rate, per-tag breakdowns
`image-to-video`	Image-to-video arena rankings	ELO, rank, CI95, appearances, win rate, per-tag breakdowns

Sources¶

This repo intentionally avoids relying on the rate-limited free AA API for daily full snapshots.

Instead it collects from Artificial Analysis public web surfaces:

llms → https://artificialanalysis.ai/leaderboards/models
text-to-image → https://artificialanalysis.ai/api/text-to-image/arena/preferences?supports_image_input=false
image-editing → https://artificialanalysis.ai/api/text-to-image/arena/preferences?supports_image_input=true
text-to-speech → https://artificialanalysis.ai/api/text-to-speech/arena/preferences
text-to-video → https://artificialanalysis.ai/api/text-to-video/arena/preferences?supports-image-input=false
image-to-video → https://artificialanalysis.ai/api/text-to-video/arena/preferences?supports-image-input=true

Structure¶

data/
├── latest.json          # → {"date": "2026-04-10", "path": "data/2026-04-10"}
├── 2026-04-10/
│   ├── _index.json      # Daily summary
│   ├── llms.json
│   ├── text-to-image.json
│   ├── image-editing.json
│   ├── text-to-speech.json
│   ├── text-to-video.json
│   └── image-to-video.json

Each data file includes:

meta.endpoint
meta.source_type
meta.source_url
meta.parser_version
meta.fetched_at
meta.model_count

Quick Access¶

## Latest pointer
curl -s https://raw.githubusercontent.com/oolong-tea-2026/artificial-analysis-leaderboards/main/data/latest.json

## Latest LLM snapshot
curl -s https://raw.githubusercontent.com/oolong-tea-2026/artificial-analysis-leaderboards/main/data/2026-04-10/llms.json

Updates¶

Data is fetched daily at 05:13 UTC via GitHub Actions.

Attribution¶

Data provided by Artificial Analysis. Please provide attribution when reusing the data. See their methodology for benchmark details.

License¶

MIT — Data attribution to Artificial Analysis required.

已知说明 (Known Issues)¶

The media arena endpoints Text-to-Image / Image Editing / Text-to-Speech now require a logged-in user key: Artificial Analysis changed the public arena API (/api/<media>/arena/preferences) so that it returns HTTP 400 {"error":"User key is required"} for unauthenticated requests. There is no public alternative endpoint (the leaderboard pages are client-rendered with no embedded data).

When the daily fetch hits this, scripts/fetch_leaderboards.py degrades gracefully instead of failing: - It falls back to the most recent historical data/<date>/<slug>.json snapshot that still contains model data. - The promoted data/<today>/<slug>.json keeps the original payload but its meta is annotated with gated: true, source: "last_known_snapshot", original_date, and a note explaining the gated API. - The daily _index.json records gated: true for those endpoints rather than only an error.

The LLM leaderboard (llms) is unaffected and is refreshed normally. Text-to-Video / Image-to-Video remain empty boards (0 models) by design. The script only exits non-zero when the critical llms endpoint itself fails.