Vector Eval - WinoGrande: leaderboard
Vector Institute Inspect Evals run of WinoGrande, testing commonsense coreference resolution on adversarially filtered pronoun disambiguation questions.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 | 93.92 | #183 |
| 2 | DeepSeek R1 | 91.79 | #251 |
| 3 | Claude 3.5 Sonnet | 90.21 | #350 |
| 4 | GPT-4o | 90.13 | #343 |
| 5 | Gemini 1.5 Pro | 87.69 | #445 |
| 6 | Llama 3.1 70B Instruct | 86.66 | #561 |
| 7 | Mistral Large 2 (Jul) | 85.48 | #484 |
| 8 | O3 Mini | 84.93 | #272 |
| 9 | Qwen3 Coder Next | 82 | #326 |
| 10 | Gemini 1.5 Flash | 78.85 | #618 |
| 11 | GPT-4o Mini | 75.3 | #599 |
| 12 | Command-R+ | 74.9 | #891 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vector-eval-winogrande · How It Works · Data refreshed daily, snapshot 2026-10-20.