Agent task completion on ClawGym-Bench benchmark leaderboard · SOTA2 Research

Agent task completion on ClawGym-Bench

76Product & Collaboration ScoreGPT-5.4

Evaluation Results

Method	Links
GPT-5.42026.04		76	66.56	73.83	76.19	68.51	82.4	73.49
Claude-4.7-Opus2026.04		73.02	74.28	80.14	80.98	81.04	82.04	77.81
GLM-5.12026.04		70.95	64.11	71.71	75.99	74.09	73.86	71.12
Claude-4.6-Sonnet2026.04		70.44	68.04	76.27	74.59	74.21	73.4	72.4
Gemini-3-Flash2026.04		68.7	68.06	68.89	78.87	80.68	84.04	73.5
Claude-4.6-Opus2026.04		68.02	63.49	73.23	74.4	73.38	82.9	71.43
Qwen3.5-Plus2026.04		67.34	66.56	74.27	73.07	71.79	72.1	70.35
Kimi-K2.62026.04		67.16	55.99	72.56	72.84	71.56	82.47	69.05
DeepSeek-V3.22026.04		64.45	64.43	71.97	72.3	66.66	70.34	67.9
MiniMax-M2.72026.04		63.98	60.11	63.96	63.42	68.72	64.1	63.72
Qwen3-235A23B2026.04		53.66	52.27	47.18	61.39	67.33	49.23	54.48
ClawGym-30A3B2026.04		52.98	50.97	64.64	61.46	57.9	56.13	56.82
ClawGym-8B2026.04		49.47	46.83	46.35	55.37	52.29	54.78	50.24
ClawGym-4B2026.04		45.21	46.6	47.51	48.99	52.25	48.27	47.73
Qwen3-30A3B2026.04		42.47	42.09	45.04	51.95	45.98	46.24	45.11
Qwen3-32B2026.04		40.68	36.21	37.84	47.16	49.29	33.11	40.32
Qwen3-8B2026.04		37.46	29.06	30.4	41.12	44.47	30.69	35.02