Benchmarks
Agentic benchmarks
These benchmarks measure how well models carry out tasks that take several steps.
We track 379 benchmarks in this topic. Main benchmarks are listed by how many models they cover, most first. Where a benchmark has separate results for a subset of its tasks, those results appear right under it. Click one for its leaderboard.
Showing 1–100 of 379 benchmarks.
- Arena Agent62 models
- Arena Agent Bash Reliability62 models
- Arena Agent Praise Complaint62 models
- Arena Agent Steerability62 models
- Arena Agent Task Outcome Explicit62 models
- Arena Agent Tool Hallucination62 models
- Terminal-Bench61 models
- Terminal-Bench 2.060 models
- Design Arena Elo (agents/mobileapps)54 models
- Arena Agent Bash Recovery Steps50 models
- Design Arena Elo (agents/fullstack)48 models
- Design Arena Elo (agents/nativeapps)42 models
- BALROG37 models
- BALROG BabaIsAI37 models
- BALROG BabyAI37 models
- BALROG Crafter37 models
- BALROG MiniHack37 models
- BALROG NetHack37 models
- BALROG TextWorld37 models
- METR Time Horizons35 models
- Agents' Last Exam32 models
- Tau2 Telecom30 models
- KiloBench Terminal-Bench29 models
- OSWorld29 models
- TAU-bench Retail25 models
- Arena ELO Image Edit - Agentic24 models
- mcp-universe-llm23 models
- TAU-bench Airline23 models
- Tau2 Retail22 models
- Design Arena Elo (agents/agon_godot)21 models