Multi-domain language model evaluation on FusionFactory full (14 tasks) (test) benchmark leaderboard · SOTA2 Research

Multi-domain language model evaluation on FusionFactory full (14 tasks) (test)

61.2Math ScoreBest LLM

Evaluation Results

Method
Best LLM2025.07	61.2	54.5	78.6	34.5	87.2	44.1	60
Thought-level FusionFactory2025.07	61.2	55	77.8	31.2	96.6	44.7	61.1
Self-Consistency2025.07	61	53.5	75.8	32.6	88.4	42	58.9
Thought-level FusionFactory2025.07	61	42.3	77.8	30.1	95.6	44.3	58.5
LLM-Blender2025.07	60.4	54.8	77.5	33.2	91.8	43.8	60.3
Zero-shot2025.07	60.2	51.7	71.5	33.4	83.4	39.8	56.7
Query-level FusionFactory2025.07	59.9	52.2	79.2	33.8	86.2	39.5	58.5
Thought-level FusionFactory2025.07	58.8	56.2	79.6	30.3	97	47.3	61.5
Thought-level FusionFactory2025.07	57.6	56	77.4	29.7	95.8	46.1	60.4
Thought-level FusionFactory2025.07	54.4	56.4	79.9	30.4	96.5	45.2	60.5
FuseLLM-adapted2025.07	54.2	45.5	79.2	28.6	84.6	41.8	55.7
Model-level FusionFactory2025.07	53	51	70.8	28.6	85.1	41	54.9
Model-level FusionFactory2025.07	52.8	43.6	80.2	29	85.1	42.5	55.5