Hands-on benchmarks, performance tests, and visual explorations for LLMs.
Test the speed, quality, and load behavior of an OpenAI-compatible endpoint.
These benchmarks are agent-first: each one has a runbook (AGENTS.md) that a coding agent can follow. You can also run them by hand.
Start here: benchmarks/README.md →
Games that compare models by how they play. Both have their own repos.
- 🏇 LLM Derby — models race when they answer trivia. It compares accuracy, speed, and cost. Play live →
- 🐍 LLM Snake Arena — two models play Snake against each other. It compares response speed, reliability, and decision-making. Play live →
Visually explore models available on Token Factory via intelligence, pricing, release timelines, and context-window trends.
Explore models → (ready to use, no API key needed)
🧪 Few testing utils — smoke-tests the Chat Completions and Responses APIs and measures streaming throughput and more.
Playable games, simulations, demos, and 3D scenes created with AI coding agents and models.
- 🎮 Quake Game — a browser-playable take on the classic game, created with coding agents and Kimi K3. Play live →