CountBenchQA
7 models
Top 5 Models Performance
| Qwen: qwen3-vl-32b-instruct | ######################################## | 93.6 |
| NVIDIA Corporation: cosmos3-super | ###################################### | 89.1 |
| NVIDIA Corporation: cosmos-reason2-32b | ##################################### | 87.5 |
| Google: Gemma 4 31B Instruct | ################################## | 79.1 |
| Google: Gemma 3 27B Pretrained | ############################# | 68 |
| Rank | Model | Score | |
|---|---|---|---|
| 🇨🇳 | 1 | Qwen: qwen3-vl-32b-instruct通义千问 | 93.6 |
| 🇺🇸 | 2 | NVIDIA Corporation: cosmos3-super | 89.1 |
| 🇺🇸 | 3 | NVIDIA Corporation: cosmos-reason2-32b | 87.5 |
| 🇺🇸 | 4 | Google: Gemma 4 31B Instruct | 79.1 |
| 🇺🇸 | 5 | Google: Gemma 3 27B Pretrained | 68 |
| 🇺🇸 | 6 | Google: gemma-3-4b-pt | 26.1 |
| 🇺🇸 | 7 | Google: gemma-3-12b-pt | 17.8 |