Agent task execution on SkillsBench 11 domains benchmark leaderboard · SOTA2 Research

Agent task execution on SkillsBench 11 domains

44.8Overall ScoreHuman

Evaluation Results

Method	Links
Human2026.06		44.8	42.5	48.6	48.3	58.2	42.5	27.5	40	53.3	0	90	40	19.8
Human2026.06		44.5	38.8	50	46.7	36.4	55	30	36	66.7	46.7	80	50	19
OPENSKILL2026.06		43.6	59.9	50	35	39.6	44.1	25	36	60	0	69.6	50	18.1
OPENSKILL2026.06		42.1	49.1	44.3	48.6	30.4	52.5	25	40	80	0	27.9	50	17.1
Skill-Creator2026.06		34.7	51.3	21.4	36.2	38.5	24.6	27.5	36	60	0	31.2	50	9.2
CoT2026.06		33.3	47.2	26.2	29.8	31.8	34.7	25	40	55.6	0	30.2	33.3	8.3
Self-Gen2026.06		32.2	48.4	31	30.3	31	20.8	29.2	26.7	33.3	11.1	30.2	33.3	7.2
Memento2026.06		30.1	34.4	31.4	35	21.8	28.8	25	32	60	0	25	30	4.6
Skill-Creator2026.06		29.2	44.4	26.2	21.9	30.9	36.8	20.8	20	22.2	0	24.3	50	4.2
No Skill2026.06		25.5	32.6	17	25.6	36.1	17.8	17.5	27.6	41.2	0	24.8	43.2	—
No Skill2026.06		25	33.2	32.9	30.4	31.3	25	0	16	0	0	29.2	30	—
AutoSkill2026.06		24.7	36	25.7	33.3	23.6	16.6	25	4	33.3	0	14.5	0	0.8
Self-Gen2026.06		23.9	37.9	16.7	31.3	27.9	18.8	16.7	13.3	11.1	0	19.8	30	1.6
CoT2026.06		23.9	34.9	17.1	30	20.4	20.4	20	16	40	0	19.2	30	1.6
Memento2026.06		15.6	19.5	14.3	13.8	18.2	12.5	12.5	26.6	22.3	0	16.5	0	9.4
AutoSkill2026.06		11.2	16.7	9.4	5.5	15.2	4.1	8.4	13.4	11	0	20	33.5	13.8