Faithful Calibration on Average across datasets (PQA, SA, SQA, HE, MMLU, SQ, MT, UM, AC, SG) benchmark leaderboard · SOTA2 Research

Faithful Calibration on Average across datasets (PQA, SA, SQA, HE, MMLU, SQ, MT, UM, AC, SG)

0.85cMFG*Qwen3-4B + RLMF + Rewr.

Evaluation Results

Method	Links
Qwen3-4B + RLMF + Rewr.2026.06		0.85	56	0.28
Llama3.1-8B-Ins + RLMF2026.06		0.84	41	0.26
Qwen3-8B + RLMF2026.06		0.83	57	0.19
Qwen3-8B + RLMF + Rewr.2026.06		0.83	57	0.19
Qwen3-4B + RLMF2026.06		0.83	56	0.28
Qwen3-1.7B + RLMF + Rewr.2026.06		0.83	50	0.23
Llama3.1-8B-Ins + RLMF + Rewr.2026.06		0.82	41	0.26
Qwen3-1.7B + RLMF2026.06		0.82	50	0.22
Llama3.1-8B-Ins + RL2026.06		0.77	40	0.2
Qwen3-4B + RL2026.06		0.76	41	0.19
Qwen3-1.7B + RL2026.06		0.75	37	0.28
Gemini-3.1-Pro2026.06		0.7	78	0.15
Llama3.1-8B-Ins + MetaFaith2026.06		0.67	28	0.36
Qwen3-8B + FUT2026.06		0.67	38	0.41
Qwen3-4B + FUT2026.06		0.67	57	0.29
Llama3.1-8B-Ins + FUT2026.06		0.66	31	0.29
Qwen3-4B + MetaFaith2026.06		0.66	53	0.3
Gemini-3-Flash2026.06		0.66	72	0.16
Qwen3-1.7B + FUT2026.06		0.64	32	0.43
Qwen3-8B + MetaFaith2026.06		0.63	51	0.29
Qwen3-1.7B + MetaFaith2026.06		0.62	45	0.3
GPT-52026.06		0.61	69	0.19
Llama3.1-8B-Ins2026.06		0.6	31	0.33
Qwen3-8B2026.06		0.54	55	0.31
Qwen3-4B2026.06		0.54	57	0.43
Qwen3-1.7B2026.06		0.52	49	0.46
Qwen3-8B + RL2026.06		0.51	59	0.26