New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Winogrande benchmark
AI model leaderboard for the Winogrande benchmark. Compare how large language models score on Winogrande, see the full ranking, and understand what this AI benchmark measures. Yi-1.5-34B currently leads with 84.9. Large-scale Winograd schema challenge — pronoun coreference resolution requiring commonsense.
Leaderboard
| # | Model | Organization | Score | Variant | Source |
|---|---|---|---|---|---|
| #1 | Yi-1.5-34B | 01.AI | 84.9 | cited-qwen2-57b | official ↗ |
| #2 | Nemotron-H 56B | NVIDIA | 84.45 | base-5-shot | official ↗ |
| #3 | Marin 8B | Marin | 84.4 | base-from-hf-readme | official ↗ |
| #4 | Nemotron-H 47B | NVIDIA | 83.9 | base-5-shot | official ↗ |
| #5 | Llama 3.1-8B | Meta AI | 82.9 | base-cited-marin | official ↗ |
| #6 | Jamba | AI21 Labs | 82.5 | cited-qwen2-57b | official ↗ |
| #7 | Mixtral 8x7B | Mistral AI | 81.9 | cited-qwen2-57b | official ↗ |
| #8 | Qwen1.5-32B | Qwen | 81.5 | cited-qwen2-57b | official ↗ |
| #9 | Phi-3.5-MoE | Microsoft | 81.3 | instruct-5-shot | official ↗ |
| #10 | Nemotron-H 8B | NVIDIA | 80.51 | base-5-shot | official ↗ |
| #11 | Qwen2-57B-A14B | Qwen | 79.5 | base-from-hf-readme | official ↗ |
| #12 | Gemma 7B | Google DeepMind | 79.01 | cited-falcon-mamba | official ↗ |
| #13 | Gemma 2 9B | Google DeepMind | 78.8 | base-cited-olmo2 | official ↗ |
| #14 | Llama 3-8B | Meta AI | 78.45 | base-cited-falcon-mamba | official ↗ |
| #15 | OLMo 2 32B | Allen Institute for AI | 78.4 | base-from-hf-readme | official ↗ |
| #16 | Mistral 7B | Mistral AI | 78.37 | v0-1-cited-falcon-mamba | official ↗ |
| #17 | Qwen2-7B | Qwen | 77 | base-from-hf-readme | official ↗ |
| #18 | GPT-4o mini | OpenAI | 76.9 | cited-phi35moe | official ↗ |
| #19 | Llama 3.1-8B | Meta AI | 76.6 | base-cited-olmo2 | official ↗ |
| #20 | Llama 2-13B | Meta AI | 74.9 | base-cited-olmo2 | official ↗ |
| #21 | Gemini 1.5 Flash (Sep 2024) | Google DeepMind | 74.7 | cited-phi35moe | official ↗ |
| #22 | Phi-2 | Microsoft | 74.4 | cited-qwen2 | official ↗ |
| #23 | Qwen2.5-7B | Qwen | 74.2 | base-cited-olmo2 | official ↗ |
| #24 | Gemma 2 9B | Google DeepMind | 74 | instruct-cited-phi35moe | official ↗ |
| #25 | Falcon Mamba | Technology Innovation Institute | 73.64 | base-from-hf-readme | official ↗ |
| #26 | DCLM 7B | Apple | 72.69 | from-hf-readme | official ↗ |
| #27 | phi-3.5-mini | Microsoft | 72.2 | cited-phi4-mini | official ↗ |
| #28 | Qwen2.5-7B | Qwen | 71.1 | cited-phi4-mini | official ↗ |
| #29 | Mistral NeMo | Mistral AI | 70.4 | instruct-cited-phi35moe | official ↗ |
| #30 | Falcon3-7B | Technology Innovation Institute | 70.4 | instruct-0-shot | official ↗ |
| #31 | Phi-4 Mini | Microsoft | 67 | instruct-5-shot | official ↗ |
| #32 | Gemma 2B | Google DeepMind | 66.8 | cited-qwen2 | official ↗ |
| #33 | Qwen2-1.5B | Qwen | 66.2 | base-from-hf-readme | official ↗ |
| #34 | Llama 3.1-8B | Meta AI | 64.7 | instruct-cited-phi35moe | official ↗ |
| #35 | Qwen2.5-3B | Qwen | 63.3 | cited-phi4-mini | official ↗ |
| #36 | Ministral 8B | Mistral AI | 59.8 | cited-phi4-mini | official ↗ |
| #37 | Qwen2-0.5B | Qwen | 56.8 | base-from-hf-readme | official ↗ |
| #38 | Llama 3.2 3B | Meta AI | 53.2 | cited-phi4-mini | official ↗ |
Frequently asked questions about Winogrande
What is the Winogrande benchmark?
Large-scale Winograd schema challenge — pronoun coreference resolution requiring commonsense.
How is the Winogrande benchmark scored?
Winogrande is scored using the accuracy 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 Winogrande?
As of the latest reported scores on GenAIList, Yi-1.5-34B achieves the highest result on Winogrande with a score of 84.9.
Is a higher Winogrande score better?
Yes. On Winogrande a higher score indicates better performance, so models near the top of the leaderboard are the strongest.