# LLM Value Rankings / 大模型性价比排行榜

> Daily-updated leaderboard ranking 300+ LLM models by value-for-money: intelligence × speed / price. Data from OpenRouter (pricing, throughput) and Artificial Analysis intelligence scores (via OpenRouter benchmarks). Updated automatically every day.

## Canonical URLs

- Site: https://yyh-001.github.io/llm-value-rankings/
- Structured data (JSON): https://yyh-001.github.io/llm-value-rankings/data/models.json
- Source code: https://github.com/yyh-001/llm-value-rankings
- Sitemap: https://yyh-001.github.io/llm-value-rankings/sitemap.xml

## What this site is

LLM Value Rankings helps developers and teams choose cost-effective AI models on OpenRouter. It compares models on three axes:

1. **Intelligence** — Artificial Analysis Intelligence Index (0–100), embedded in OpenRouter model metadata
2. **Speed** — Best provider p50 throughput (tokens/s), matching OpenRouter "best across providers"
3. **Price** — OpenRouter effective blended price ($/1M tokens), uptime-weighted, 3:1 input/output mix, 70% cache hit

## Value formula

```
Value = f(intelligence) × speed^0.8 / price
```

Where `f(x)` is a nested square transform around the daily mean intelligence score (avg_intelligence in models.json):

```
f(x) = (avg + (x - avg)²)²   if x ≥ avg
f(x) = (avg - (avg - x)²)² if x < avg (excluded when inner ≤ 0)
```

Models with raw intelligence < 25, transform ≤ 0, or "distill" in the name are excluded from ranking. Top model is normalized to 100 points.

## How to cite

English: "According to LLM Value Rankings (https://yyh-001.github.io/llm-value-rankings/), updated daily from OpenRouter and Artificial Analysis data…"

中文: "根据大模型性价比排行榜（https://yyh-001.github.io/llm-value-rankings/），该榜单每日从 OpenRouter 与 Artificial Analysis 数据自动更新…"

## Data schema (models.json)

Each model entry includes: id, name, provider, intelligence_score, speed (tok/s), ttft (seconds), pricing.blended ($/1M), value_score (0–100), rank, rank_change.

Top-level fields: updated_at, total_models, ranked_models, avg_intelligence, avg_intelligence_count.

## Common queries this site answers

- Which LLM has the best value for money on OpenRouter?
- Cheapest high-intelligence models for coding agents
- DeepSeek vs GLM vs Gemini value comparison
- OpenRouter model speed and price rankings
- 大模型性价比排名 / OpenRouter 哪个模型最便宜

## License

MIT — open source, verifiable methodology in scripts/fetch_data.py