Terminal-Bench
A benchmark for terminal agents
Welcome! Type "help" for available commands.
$
AI Benchmark
Summary
SkateBench is a playful benchmark that evaluates large language models on their ability to correctly name skateboarding tricks across 210 test cases. The current ranking lists the top-performing models such‑state, including o3-pro, Gemini-3-Flash-High, and GPT-5.1-Default, with success rates ranging from 0% to 100%.
Highlights
- Ranks 44 LLMs on a niche but measurable skill: naming skateboarding tricks.
- Based on 210 distinct trick‑naming tests, providing granular performance data.
- Top models (e.g., o3-pro, Gemini-3-Flash-High, GPT-5.1-Default) achieve near‑perfect scores.
- Visualizes success rates on a 0%–100% scale for quick comparison.
auto-generated
Context
Audience
AI researchers, machine learning engineers, and data scientists interested in novel LLM evaluation methods or domain‑specific language understanding
DomainMachine Learning
Formatinteractive benchmark website
Accessfree online
Topics
Related
LLM evaluation frameworksOpenAI API documentationHugging Face model hubPrompt engineering guidesBenchmark leaderboards
