Automated Student Response Grading on Beetle benchmark leaderboard · SOTA2 Research

Automated Student Response Grading on Beetle

0.963Thetaopenchat/openchat-3.5-0106

Evaluation Results

Method
openchat/openchat-3.5-01062026.04	0.963	57.7	49.4
Qwen/Qwen3.5-9B2026.04	0.653	51.8	39.5
google/gemma-3-4b-it2026.04	0.515	49.6	43.4
google/gemma-3-12b-it2026.04	0.487	49.4	46.3
Qwen/Qwen3.5-4B2026.04	0.404	47	31.4
microsoft/Phi-3-medium-4k-instruct2026.04	0.33	46.4	37.5
microsoft/Phi-3-mini-4k-instruct2026.04	0.197	45	41.4
Qwen/Qwen3.5-0.8B2026.04	0.109	41.2	19.3
microsoft/Phi-4-mini-instruct2026.04	0.084	41.6	25.3
google/gemma-3-1b-it2026.04	-0.075	39.1	27.3
meta-llama/Llama-3.1-8B-Instruct2026.04	-0.125	38.6	33.8
meta-llama/Llama-3.2-1B-Instruct2026.04	-0.342	33.1	18.8
Qwen/Qwen3.5-2B2026.04	-0.463	32.2	17.2
mistralai/Ministral-3-14B-Instruct-25122026.04	-0.645	29.6	24.6
mistralai/Ministral-3-3B-Instruct-25122026.04	-0.645	29.8	26.4
mistralai/Ministral-3-8B-Instruct-25122026.04	-0.649	29.7	28.2
meta-llama/Llama-3.2-3B-Instruct2026.04	-0.798	26.6	16.1