Qwen
Benchmarks
TaskDatasetSOTA resultTrendResults
Qwen3-30B-A3B
16.2Memory Usage (GB)
3
Qwen2.5-7B
5.6Mean PPL
3
Qwen 0.5B 2.5
13.39Mean Perplexity (PPL)
3
Qwen2.5-7B (test)
71.533Encoding Throughput (Token/s)
3
Qwen 7B Nano 8G 2.5 (inference)
57.7TPOT (ms)
3
Qwen-2.5-7B on Jetson Orin Nano 8G (512 prefill, 512 generation tokens) (inference)
143.12Energy per Request
3
Qwen3-8B vLLM inference
1.732Model Load Time (s)
3
Qwen3-4B vLLM (inference)
1.101Model Load Time (s)
3
Qwen3-0.6B vLLM inference
0.336Model Load Time (s)
3
Qwen2.5-72B-Instruct Long-Context Efficiency Workload
4.76Speedup (8K)
2
Qwen3-8B
0MSE
2
Qwen 4B 3.5
6.74MSE
2
Qwen2-VL vid-only workload
86.7Pairwise Accuracy
2
Qwen MoE workload 1.5
94.8Pairwise Accuracy
2
Qwen3-235B-A22B (train)
872.4Average Optimizer State Memory (MB)
2
Qwen3-235B workload (10,000 req/s on AMD MI300X)
137Nodes
2
Qwen 7B-Instruct LoRA 2.5 (train)
79Ball Acceptance Rate
2
Mini-Qwen3 (val)
1.763Validation Loss
2
Qwen-3 30B-A3B (64 x B200-8)
4.1Iteration Time (s)
2
Qwen-3 30B-A3B (tpu-v5p-1024)
12.86Iteration Time (s)
2
Qwen2.5 1.5B-family
0.99TPR
2
Qwen3-RM
158W-L Differential
1
Qwen-3 family
98TPR
1