New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
ArenaHard v2 benchmark
AI model leaderboard for the ArenaHard v2 benchmark. Compare how large language models score on ArenaHard v2, see the full ranking, and understand what this AI benchmark measures. GPT-5 currently leads with 82.91. Arena-Hard v2 — automatic preference benchmark of 500 hard prompts judged by an LLM, used to estimate model rankings against the LMArena leaderboard.
Leaderboard
| # | Model | Organization | Score | Variant | Source |
|---|---|---|---|---|---|
| #1 | GPT-5 | OpenAI | 82.91 | thinking-high-win-rate-cited-ring-1t | official ↗ |
| #2 | Ring-1T | Ant Group | 81.59 | thinking-win-rate | official ↗ |
| #3 | Qwen3-235B-A22B-Thinking (Jul 2025) | Qwen | 80.18 | thinking-win-rate-cited-ring-1t | official ↗ |
| #4 | gemini-2.5-pro | Google DeepMind | 79.2 | win-rate-cited-ring-1t | official ↗ |
| #5 | Ling-1T | Ant Group | 75.83 | win-rate | official ↗ |
| #6 | gemini-2.5-pro | Google DeepMind | 74.46 | lowthink-win-rate-cited-ling-1t | official ↗ |
| #7 | Kimi K2 | Moonshot | 69.88 | instruct-0905-win-rate-cited-ling-1t | official ↗ |
| #8 | GPT-5 | OpenAI | 65.06 | main-win-rate-cited-ling-1t | official ↗ |
| #9 | DeepSeek-V3.1-Terminus | DeepSeek | 63.24 | non-thinking-win-rate-cited-ling-1t | official ↗ |
| #10 | DeepSeek-V3.1-Terminus | DeepSeek | 60.27 | thinking-win-rate-cited-ring-1t | official ↗ |
| #11 | Mistral Small 3.2 | Mistral AI | 43.1 | instruct-from-hf-readme | official ↗ |
| #12 | Mistral Small 3.1 | Mistral AI | 19.56 | instruct-cited-msmall32 | official ↗ |
Frequently asked questions about ArenaHard v2
What is the ArenaHard v2 benchmark?
Arena-Hard v2 — automatic preference benchmark of 500 hard prompts judged by an LLM, used to estimate model rankings against the LMArena leaderboard.
How is the ArenaHard v2 benchmark scored?
ArenaHard v2 is scored using the win_rate metric, where a higher score is better. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.
Which AI model scores highest on ArenaHard v2?
As of the latest reported scores on GenAIList, GPT-5 achieves the highest result on ArenaHard v2 with a score of 82.91.
Is a higher ArenaHard v2 score better?
Yes. On ArenaHard v2 a higher score indicates better performance, so models near the top of the leaderboard are the strongest.