Tokyo LLM Benchmarks: Open-source vs. Closed-source Large Language Models Real-time Comparison Visualization Dashboard

Core Project Introduction

Tokyo LLM Benchmarks is an interactive dashboard with a Tokyo night view glassmorphism style that real-time tracks the performance gap between open-source and closed-source large language models, showing Elo score comparisons between open-source models like GLM-5.2 Max and DeepSeek V4 Pro and closed-source giants like GPT-5.5 and Claude Fable 5.

Basic Information

• Original Author/Maintainer: David Linacre (GitHub: @LIN4CRE)
• Source Platform: GitHub
• Original Title: LLM-Tokyo-Benchmarks
• Original Link: https://github.com/LIN4CRE/LLM-Tokyo-Benchmarks
• Release Time: July 14, 2026
• License: MIT
• Online Demo: https://LIN4CRE.github.io/LLM-Tokyo-Benchmarks/

In the field of large language models in 2026, the performance gap between open-source models and closed-source commercial models has narrowed significantly. Open-source architectures like GLM-5.2 Max and DeepSeek V4 Pro are competing head-on with closed-source giants like GPT-5.5 and Claude Fable 5. This fierce competition makes tracking model performance more important and difficult, so the Tokyo LLM Benchmarks project was born as a real-time interactive visualization dashboard to help understand the competition between open-source and closed-source models.

Tokyo LLM Benchmarks adopts the "Tokyo Night View" glassmorphism design theme, using neon-style visuals to distinguish model types: open-source models use glowing neon green, closed-source models use high-contrast neon orange, which is both aesthetically pleasing and easy for quick identification. The overall design creates an immersive Tokyo night view atmosphere, enhancing the user experience.

Key Features

1. Exquisite Dark Glass UI: CSS4 glassmorphism effects + dynamic background light spots, clear and easy to read.
2. Interactive Data Visualization: Uses Chart.js to implement 24-hour rolling average Elo score tracking, showing performance change trends.
3. Simulated Terminal Synchronization: Simulates connecting to arena.ai and llm-stats.com data sources via the "Scrape and Sync" button; the terminal output stream adds a technical feel.
4. Categorized Model Display:
   • Closed-source Giants: Claude Fable 5 (Elo1509, $10 per million tokens), GPT-5.5 Pro (Elo1506, $30 per million tokens), Gemini3.1 Pro (Elo1485, $2 per million tokens)
   • Open-source Heroes: GLM-5.2 Max (Elo1469, Free/Self-hosted), Mimo V2.5 Pro (Elo1466, Free/Self-hosted), DeepSeek V4 Pro (Elo1457, Free/Self-hosted)

The project's tech stack is concise and efficient:

• HTML5: Semantic structure, ensuring accessibility and SEO-friendliness.
• CSS4: Custom properties, Flex/Grid layout, advanced background blur filters, keyframe animations.
• Native JavaScript (ES6+): DOM operations and terminal logic, no heavy frameworks.
• Chart.js: Responsive Canvas rendering, lightweight and powerful.
• GitHub Actions: Continuous integration/deployment; pushing to the main branch automatically deploys to GitHub Pages. No complex build steps are needed—just open the HTML file to run, lowering the barrier to use.

1. Model Selection Reference: Open-source models (e.g., GLM-5.2 Max) have Elo scores close to closed-source models and zero cost, providing a decision-making basis for teams with limited budgets.
2. Open-source Community Confidence: Shows the possibility of open-source models catching up with closed-source ones, inspiring developers to participate in open-source model optimization.
3. Technical Trend Insight: 24-hour rolling data tracking helps observe short-term fluctuations and long-term trends of model performance, providing data sources for researchers.

Limitations

1. Data sources are based on simulation logic, not real-time API access.
2. Elo scores cannot fully represent performance in actual application scenarios.
3. Data updates depend on manual scraping by maintainers.

Future Improvement Directions

• Integrate real-time APIs to achieve true real-time updates.
• Add evaluation dimensions such as inference speed, memory usage, and multilingual support.
• Add historical data comparison functionality.
• Support user-customized model comparisons.

Tokyo LLM Benchmarks combines aesthetic value and technical practicality, showing the strong performance of open-source models in 2026 and providing developers with an intuitive model comparison tool. The project is open-source (under MIT license), and community contributions are welcome. You can visit the online demo to experience it, or clone the repository to run it locally.