Zero-shot Reasoning and Language Modeling on Qwen1.5-MoE-A2.7B Evaluation Suite benchmark leaderboard · SOTA2 Research

Zero-shot Reasoning and Language Modeling on Qwen1.5-MoE-A2.7B Evaluation Suite

6.79PPLbf16 baseline

Evaluation Results

Method	Links
bf16 baseline2026.06		6.79	43.86	68.98	77.17	71.57	64.56	80.36	69.22	67.96
SliM-LLM2026.06		6.88	44.62	68.94	76.55	71.18	65.07	80.47	69.14	68
MoE-Salient-AQ2026.06		6.89	44.37	69.91	76.92	71.38	64.27	80.2	68.82	67.98
MoE-Salient-AQ2026.06		6.97	45.31	70	76.75	70.64	63.71	80.36	68.9	67.95
MxMoE2026.06		7.09	44.03	68.01	76.29	69.18	62.95	79.49	68.11	66.87
MxMoE2026.06		7.1	40.87	68.06	76.21	69.69	62.33	80.69	68.9	66.68
MoE-Salient-AQ2026.06		7.16	43.69	67.55	75.94	69.36	63.15	79.22	66.61	66.5
SliM-LLM2026.06		7.27	40.36	64.86	74.94	69.47	63.79	78.94	65.75	65.44
MxMoE2026.06		7.43	43.43	67.21	75.17	68.41	60.76	79.49	67.64	66.02
MoE-Salient-AQ2026.06		7.81	40.1	62.46	73.97	67.03	60.28	79.49	63.85	63.88
MxMoE2026.06		8.75	37.37	59.85	70.57	60.86	55.06	77.91	62.12	60.53