Claw-Eval

29 models

ClawEval is a cutting-edge AI benchmark consisting of 300 human-verified tasks across 9 categories, specifically designed to stress-test autonomous LLM agents in real-world workflows. Its defining feature is full-trajectory auditing: rather than just checking the final output, the system evaluates the models on execution traces, safety, and robustness across 2,159 fine-grained rubrics in a sandboxed environment, making this dataset the ultimate gold standard for measuring practical agentic capabilities.

Top 5 Models Performance

Ornith: ornith-1.5-397b	########################################	81.4
Ornith: ornith-1.5-122b	#######################################	79.4
nanbeige: nanbeige4.3-33b	######################################	78.3
Ornith: ornith-1.0-397b	######################################	77.1
Anthropic: claude-opus-4.5	######################################	76.6

	Rank	Model	Score
❓	1	Ornith: ornith-1.5-397b	81.4
❓	2	Ornith: ornith-1.5-122b	79.4
❓	3	nanbeige: nanbeige4.3-33b	78.3
❓	4	Ornith: ornith-1.0-397b	77.1
🇺🇸	5	Anthropic: claude-opus-4.5	76.6
❓	6	xingchen-agi: xing4.0-29b-a4b	76.55
🇨🇳	7	MiniMax: MiniMax M3稀宇科技	74.5
❓	8	Ornith: ornith-1.5-35b-a3b	72.5
🇨🇳	9	Qwen: qwen3.6-27b通义千问	72.4
❓	10	Ornith: ornith-1.0-35b	69.8
🇺🇸	11	LiquidAI: lfm-3-1.7b	68.64
🇨🇳	12	Tencent: hy3腾讯控股有限公司	68.5
❓	13	Ornith: ornith-1.5-9b	66.5
🇨🇳	14	Qwen: Qwen3.5 27B通义千问	64.3
🇨🇳	15	Xiaomi: MiMo-V2.5 Pro小米集团	64
🇨🇳	16	inclusionAI: ring-2.6-1t	63.82
❓	17	Ornith: ornith-1.0-9b	63.1
🇨🇳	18	Qwen: qwen3.7-plus通义千问	62.7
🇨🇳	19	Z.ai: GLM 5.1北京智谱华章科技有限公司	62.7
🇨🇳	20	Xiaomi: MiMo-V2.5小米集团	62.1
🇨🇳	21	Moonshot AI: Kimi K2.6月之暗面	61.5
🇨🇳	22	DeepSeek: DeepSeek V4 Pro深度求索	58.4
🇨🇳	23	DeepSeek: DeepSeek V4 Flash深度求索	57.8
🇨🇳	24	Qwen: qwen3.5-397b-a17b通义千问	57.8
🇨🇳	25	Qwen: qwen3.8-27b通义千问	57.4
🇨🇳	26	Moonshot AI: Kimi K2.5月之暗面	52.8
🇨🇳	27	MiniMax: minimax-m2.7稀宇科技	49.7
🇨🇳	28	DeepSeek: deepseek-v3.2深度求索	42.2
🇺🇸	29	NVIDIA Corporation: nvidia-nemotron-3-super-120b-a12b	6.8