Benchmark
Arena Hard Benchmark
What is Arena Hard?
Arena-Hard-Auto is an automatic evaluation benchmark for instruction-tuned LLMs consisting of 500 challenging real-world prompts curated by BenchBuilder. It…
- Modality
- text
- Evaluates
- reasoning, general, creativity, writing
- Openness
- unknown
- Importer
- LLM Stats
- Reported scores
- 26
Source provenance
- Original evidence https://api.zeroeval.com/leaderboard/benchmarks/arena-hard?top_n=500
Arena Hard paper and code
Arena Hard results and reported scores
LLM Stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| DeepSeek-V2.5 | DeepSeek | 0.762 | 2024-05-08 | evidence |
| Granite 3.3 8B Base | IBM | 0.5756 | 2025-04-16 | evidence |
| Granite 3.3 8B Instruct | IBM | 0.5756 | 2025-04-16 | evidence |
| IBM Granite 4.0 Tiny Preview | IBM | 0.267 | 2025-05-02 | evidence |
| Jamba 1.5 Large | AI21 Labs | 0.654 | 2024-08-22 | evidence |
| Jamba 1.5 Mini | AI21 Labs | 0.461 | 2024-08-22 | evidence |
| Llama-3.3 Nemotron Super 49B v1 | NVIDIA | 0.883 | 2025-03-18 | evidence |
| MiniStral 3 (14B Instruct 2512) | Mistral | 0.551 | 2025-12-04 | evidence |
| Ministral 3 (3B Instruct 2512) | Mistral | 0.305 | 2025-12-04 | evidence |
| Ministral 3 (8B Instruct 2512) | Mistral | 0.509 | 2025-12-04 | evidence |
| Ministral 8B Instruct | Mistral | 0.709 | 2024-10-16 | evidence |
| Mistral Large 3 | Mistral | 0.551 | 2025-09-01 | evidence |
| Mistral Small 3 24B Instruct | Mistral | 0.876 | 2025-01-30 | evidence |
| Mistral Small 3.2 24B Instruct | Mistral | 0.431 | 2025-06-20 | evidence |
| Mistral Small 4 | Mistral | 0.583 | 2026-03-16 | evidence |
| Phi 4 | Microsoft | 0.754 | 2024-12-12 | evidence |
| Phi 4 Mini | Microsoft | 0.328 | 2025-02-01 | evidence |
| Phi 4 Reasoning | Microsoft | 0.733 | 2025-04-30 | evidence |
| Phi 4 Reasoning Plus | Microsoft | 0.79 | 2025-04-30 | evidence |
| Phi-3.5-mini-instruct | Microsoft | 0.37 | 2024-08-23 | evidence |
| Phi-3.5-MoE-instruct | Microsoft | 0.379 | 2024-08-23 | evidence |
| Qwen2.5 72B Instruct | Qwen | 0.812 | 2024-09-19 | evidence |
| Qwen2.5 7B Instruct | Qwen | 0.52 | 2024-09-19 | evidence |
| Qwen3 235B A22B | Qwen | 0.956 | 2025-04-29 | evidence |
| Qwen3 30B A3B | Qwen | 0.91 | 2025-04-29 | evidence |
| Qwen3 32B | Qwen | 0.938 | 2025-04-29 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.