WinoGrande
Pronoun-resolution commonsense questions. A 2019 benchmark, saturated for current models; kept for history.
As of 19 Sept 2026, Llama 3.1 405B leads WinoGrande on BenchLeader with 89.2%, ahead of Claude 3 Opus at 88.5%, across 74 model configurations with a published result.
- Published by
- Epoch AI Benchmarking Hub
- Category
- Reasoning
- Index weight
- Reference only
- Models
- 74
- Data as of
- 19 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
Pronoun-resolution commonsense questions.
How it is scored
Accuracy, as compiled by Epoch AI from published results.
What to keep in mind
Saturated: frontier models score near the ceiling, so it separates only older or smaller models, and results are largely self-reported by developers.
- 1Llama 3.1 405B89.2%
- 2Claude 3 Opus88.5%
- 3GPT 4 031487.5%
- 4Falcon-180B87.1%
- 5Deepseek v286.3%
- 6DeepSeek V385.2%
- 7Palm 540B85.1%
- 8DeepSeek-Coder-V2 236B83.7%
- 9Llama 3-70B83.5%
- 10PaLM 2-L83.0%
- 11Qwen2.5 72B82.3%
- 12davinci-00281.6%
- 13GPT-3.5 Turbo (older v0613)81.6%
- 14Phi 3 Small81.5%
- 15phi-3-medium 14B (medium)81.5%
74 of 74
| # | ||||
|---|---|---|---|---|
| 1 |  Llama 3.1 405BMetaopen | 89.2% | 43.5 | 2024-07-23 |
| 2 |  Claude 3 OpusAnthropic | 88.5% | 39.6 | 2024-02-29 |
| 3 |  GPT 4 0314OpenAI | 87.5% | 43.9 | 2023-03-14 |
| 4 |  Falcon-180BTII | 87.1% | – | 2023-09-06 |
| 5 |  Deepseek v2DeepSeekopen ↗ | 86.3% | – | 2024-05-07 |
| 6 | DeepSeek V3DeepSeekopen | 85.2% | 44.8 | 2024-12-26 |
| 7 |  Palm 540BGoogle | 85.1% | – | 2022-04-04 |
| 8 | DeepSeek-Coder-V2 236BDeepSeek | 83.7% | – | 2024-06-17 |
| 9 | Llama 3-70BMetaopen | 83.5% | 39.2 | 2024-04-18 |
| 10 | PaLM 2-LGoogle | 83.0% | – | 2023-05-17 |
| 11 |  Qwen2.5 72BAlibabaopen | 82.3% | 42.7 | 2024-09-19 |
| 12 | davinci-002OpenAI | 81.6% | – | 2022-03-15 |
| 13 | GPT-3.5 Turbo (older v0613)OpenAI | 81.6% | – | 2023-06-13 |
| 14 |  Phi 3 SmallMicrosoft | 81.5% | – | 2024-04-23 |
| 15 | phi-3-medium 14BmediumMicrosoft | 81.5% | – | 2024-04-23 |
| 16 | Qwen2.5-Coder 32B InstructAlibabaopen ↗ | 80.8% | 42.6 | 2024-09-18 |
| 17 | Llama 2-70BMetaopen ↗ | 80.2% | 34.9 | 2023-07-18 |
| 18 | PaLM 2-MGoogle | 79.2% | – | 2023-05-17 |
| 19 | Gemma 7BGoogle | 79.0% | – | 2024-02-21 |
| 20 |  Megatron-Turing NLG 530BNVIDIA | 78.9% | – | 2021-10-11 |
| 21 | Falcon 2 11BTII | 78.3% | – | 2024-05-09 |
| 22 | Nemotron-4 15BNVIDIA | 78.0% | – | 2024-02-26 |
| 23 | PaLM 2-SGoogle | 77.9% | – | 2023-05-17 |
| 24 | InstructGPT 175BOpenAI | 77.7% | – | 2022-01-27 |
| 25 |  Mixtral 8x7BMistral AIopen | 77.2% | 32.1 | 2023-12-11 |
| 26 | LLaMA-65BMetaopen | 77.0% | – | 2023-02-24 |
| 27 | PaLM 62BGoogle | 77.0% | – | 2022-04-04 |
| 28 | Falcon-40BTII | 76.9% | – | 2023-03-15 |
| 29 | Qwen2.5-Coder-14BAlibaba | 76.8% | – | 2024-09-18 |
| 30 | Llama 2-34BMeta | 76.7% | – | 2023-07-18 |
| 31 | LLaMA-33BMeta | 76.0% | – | 2023-02-24 |
| 32 | Llama 3-8BMetaopen | 75.7% | 35.6 | 2024-04-18 |
| 33 | Mistral 7B v0.3Mistral AIopen | 75.3% | 32.1 | 2023-09-27 |
| 34 | Claude 3 SonnetAnthropic | 75.1% | 38.6 | 2024-02-29 |
| 35 | Claude 3 HaikuAnthropic | 74.2% | 37.6 | 2024-03-07 |
| 36 | Phi-1.5Microsoft | 73.4% | – | 2023-09-11 |
| 37 | LLaMA-13BMeta | 73.0% | – | 2023-02-24 |
| 38 |  Yi-9B01.AI | 73.0% | – | 2024-03-01 |
| 39 | Qwen2.5-Coder 7B InstructAlibabaopen ↗ | 72.9% | – | 2024-09-18 |
| 40 | DeepSeek-Coder-V2-Lite-BaseDeepSeek | 72.9% | – | 2024-06-13 |
| 41 | Llama 2-13BMetaopen ↗ | 72.8% | 36.5 | 2023-07-18 |
| 42 | Yi 6B01.AI | 71.3% | – | 2023-11-02 |
| 43 |  MPT-30BMosaicML | 71.0% | – | 2023-06-22 |
| 44 | Phi 3 MiniMicrosoftopen | 70.8% | 36.2 | 2024-04-23 |
| 45 |  vicuna-13b-v1.1LMSYS | 70.8% | – | 2023-04-12 |
| 46 | LLaMA-7BMeta | 70.1% | – | 2023-02-24 |
| 47 | Llama 2-7BMetaopen ↗ | 69.2% | 34.2 | 2023-07-18 |
| 48 | GPT 3.5 Turbo 1106OpenAI | 68.8% | 39.2 | 2023-11-06 |
| 49 | MPT-7BMosaicML | 68.6% | – | 2023-05-05 |
| 50 | Qwen2.5-Coder-3BAlibaba | 67.4% | – | 2024-09-18 |
| 51 | Falcon-7BTII | 67.2% | – | 2023-04-24 |
| 52 | open_llama_7bUnknown | 67.0% | – | 2023-06-07 |
| 53 | GPT-NeoX-20BEleutherAI | 66.1% | – | 2022-04-07 |
| 54 |  INTELLECT-1Prime Intellect | 65.8% | – | 2024-11-29 |
| 55 | Gemma 2BGoogle | 65.4% | – | 2024-02-21 |
| 56 |  XGen-7BSalesforce | 64.9% | – | 2023-06-27 |
| 57 | opt-13bUnknown | 64.7% | – | 2022-05-11 |
| 58 | GPT-J-6BEleutherAI | 64.5% | – | 2021-08-05 |
| 59 | StarCoder 2 15BNVIDIA | 64.3% | – | 2024-02-20 |
| 60 | RedPajama-INCITE-7B-BaseUnknown | 63.8% | – | 2023-05-04 |
Cite as: BenchLeader, “WinoGrande leaderboard”, https://www.benchleader.com/benchmarks/winogrande, data as of 19 Sept 2026.
WinoGrande: questions
- What does WinoGrande measure?
- Pronoun-resolution commonsense questions. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads WinoGrande?
- Llama 3.1 405B leads WinoGrande with 89.2% as of 19 Sept 2026, ahead of Claude 3 Opus at 88.5%.
- How many models have WinoGrande results?
- 74 model configurations have a WinoGrande result on BenchLeader, all taken from Epoch AI Benchmarking Hub.
- Who runs WinoGrande and how often is it updated?
- WinoGrande is published by Epoch AI Benchmarking Hub. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does WinoGrande count toward the BenchLeader Index?
- No. WinoGrande is shown for reference but left out of the composite index.