LLM Value Rankings

Find the best value AI models — intelligence per dollar, ranked daily.

🌐 Live site: https://yyh-001.github.io/llm-value-rankings/

English · 简体中文

🌐 Live Site · 🐛 Report Bug · ✨ Request Feature

---

Overview

LLM Value Rankings compares 300+ LLMs by combining:

• Intelligence — Artificial Analysis Intelligence Index (OpenRouter embedded benchmarks)
• Speed — output tokens per second (uptime-weighted average across providers)
• Price — blended input/output cost from OpenRouter, plus 7 channel price sources for comparison

Beyond OpenRouter pay-as-you-go rates, the site ingests OpenCode Go, Command Code GOAT, DeepSeek official API, GPT Plus, MiniMax Token Plan, and MiMo Token Plan pricing. Model detail pages show all channels side by side. Rank by lowest channel price is off by default; turn it on to recompute value scores and sort order from each model’s best available rate.

The result is a daily-updated leaderboard that answers one question: how much AI capability do you get per dollar?

Static site · zero backend · deploy with GitHub Pages in one minute.

---

Features

	Feature	Description
📊	Smart ranking	f(Intelligence) × Speed^0.8 / Price, min score 25
💰	Multi-channel pricing	OpenCode, Command Code, DeepSeek API, GPT Plus, MiniMax Token Plan, MiMo Token Plan + OpenRouter; detail modal sorted by lowest price
🔄	Lowest-channel ranking	Off by default: turn on to recompute value score, price column, and sort from each model’s cheapest channel (toggle persists)
📋	Channel summary	Collapsible panel next to filters: channel count, plan count, models covered
📉	Pareto chart	Top 30 intelligence vs price; hover shows which channel supplied the displayed price
📈	Day-over-day delta	See rank changes vs yesterday (↑2 / ↓1 / NEW)
🏅	Top 3 podium	Highlight the best value models on the homepage
🔍	Live search	Filter by model name or ID
🌍	Bilingual UI	Chinese / English toggle
🌙	Dark mode	System-aware theme with manual override
🎨	Theme variants	Classic, SpaceX, Apple, EVA, Minimal, and more
⭐	GitHub Star	One-click star from the header
📱	Responsive	Card layout on mobile, table on desktop
🤖	Auto-updated	GitHub Actions fetches fresh data daily

---

Quick Start

Preview locally

git clone https://github.com/yyh-001/llm-value-rankings.git
cd llm-value-rankings

python -m http.server 8080
# or: npx serve .

Open http://localhost:8080.

Deploy to GitHub Pages

1. Fork this repository
2. Go to Settings → Pages
3. Source: Deploy from a branch
4. Branch: main · Folder: / (root)
5. Visit https://<your-username>.github.io/llm-value-rankings/

Update data manually

pip install -r scripts/requirements.txt
python scripts/fetch_data.py
git add data/models.json data/rank_history.json data/coding_plans.json
git commit -m "chore: update model data"
git push

Or trigger the Update Model Data workflow from the Actions tab.

---

Methodology

Value score

Value = f(Intelligence) × Speed^0.8 / Price

f(x) is a nested square transform around the mean intelligence score:

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

This rewards models above the average more than a plain square, without the harsh spread of a pure 4th power. The UI shows the raw intelligence score; ranking uses the transformed value.

Metric	Source	Notes
Intelligence	OpenRouter embedded AA intelligence_index	Artificial Analysis Intelligence Index, 0–100
Speed	OpenRouter Endpoints API + RSC page scrape	Uptime-weighted average of provider p50 throughput
TTFT	OpenRouter Endpoints API	Time-to-first-token p50 in seconds
Price	OpenRouter	Uptime-weighted effective price: 3:1 input/output token mix, 95% cache-hit on input ($/1M)

Score scale (0–100)

After computing the raw value above, scores are normalized to a 0–100 scale. The top-ranked model is always 100; others are proportional.

Exclusion rules

Models are excluded from ranking when:

• Raw intelligence score is below 25
• The transformed capability score is ≤ 0 (inner term below zero for below-average models)
• The model name contains distill

Ranking rules

• Text-output LLMs only (image-generation models excluded via OpenRouter output_modalities)
• Requires both intelligence score and pricing data
• Models with distill in the name are excluded
• Sorted by value score descending

---

Channel pricing

By default the leaderboard ranks by OpenRouter blended price. Turn on Rank by lowest channel price beside the search bar to use each model’s cheapest channel.

Channels (7)

Channel	Type	Notes
OpenRouter	Pay-as-you-go API	Uptime-weighted blended price from models.json
OpenCode Go	$10/mo subscription	Full catalog from official docs, matched to ranked models by slug
Command Code GOAT	$10/mo subscription	GOAT plan table parsed and matched by label/slug
DeepSeek official API	Pay-as-you-go	24h peak/off-peak weighted average (7h peak + 17h off-peak CST)
GPT Plus	Codex subscription	Estimated $/M from OpenAI Codex pricing and weekly quotas
MiniMax Token Plan	MiniMax subscription	Individual monthly Plus ¥49 / Max ¥119 / Ultra ¥469, official M3 token pool; M3 and M2.7 share quota
MiMo Token Plan	Xiaomi subscription	Monthly Lite ¥39 / Standard ¥99 / Pro ¥329 / Max ¥659; Credits/token converted with agent mix (not 1:1)

All channel rates are normalized to ¥/M tokens (95% cache hit, 3:1 input/output) and shown in the model detail modal, sorted cheapest first.

Matching & scope

• Channel prices attach only to ranked models (rank set); vendor slugs/labels are matched strictly to OpenRouter model IDs
• Zhipu GLM Coding Plan and Kimi official membership/API plans are not included (credit/membership billing differs from per-token leaderboard pricing)
• Generated by scripts/fetch_coding_plans.py into data/coding_plans.json

---

Project Structure

llm-value-rankings/
├── index.html                  # Main page (classic theme)
├── spacex.html / apple.html …  # Alternate theme entry points
├── css/style.css               # Styles (light/dark, multi-theme)
├── js/
│   ├── app.js                  # Core logic (rankings, channels, modal)
│   ├── pareto-chart.js         # Pareto frontier chart
│   └── i18n.js                 # Internationalization
├── data/
│   ├── models.json             # Model data (auto-updated)
│   ├── coding_plans.json       # Channel plans & per-model index (auto-updated)
│   └── rank_history.json       # Daily rank snapshots
├── scripts/
│   ├── fetch_data.py           # Model fetch & ranking
│   ├── fetch_coding_plans.py   # Channel plan scrape & matching
│   ├── recalc_scores.py        # Offline value-score recalc (keeps speed as-is)
│   └── requirements.txt
└── .github/workflows/
    └── update-data.yml         # Daily cron job

---

Data Sources

Data	Provider	Method
Pricing & model list	OpenRouter API	REST
Intelligence	OpenRouter embedded AA benchmarks	intelligence_index field on model objects
Speed & TTFT	OpenRouter	Endpoints API + RSC page scrape
Channel plan pricing	OpenCode / Command Code / OpenAI / DeepSeek docs	fetch_coding_plans.py scrape & estimates

---

Tech Stack

Layer	Stack
Frontend	HTML · CSS · Vanilla JS
Data	Static JSON
Pipeline	Python 3.11 · GitHub Actions
Hosting	GitHub Pages

No build step. No framework. No database.

---

Contributing

Contributions are welcome! Here are good starting points:

1. Add model mappings in scripts/fetch_data.py
2. Improve UI / i18n copy
3. Refine the ranking algorithm
4. Fix bugs via Issues

# Fork → branch → commit → pull request
git checkout -b feat/your-feature

---

License

MIT © 2026