Metacognitive Ability on TriviaQA benchmark leaderboard · SOTA2 Research

Metacognitive Ability on TriviaQA

1.597d'type2Mistral-Instruct

Evaluation Results

Method	Links
Mistral-Instruct2026.03		1.597	—	42.7	—	—	—	1.361	0.852	0.765
Llama-3-Base2026.03		1.407	—	42.8	—	—	—	1.474	1.048	0.952
Llama-3-Instruct2026.03		1.386	—	54.3	—	—	—	1.362	0.983	0.886
Qwen2.5 3B ESMA2026.02		1.02	69.59	51.2	51.41	29.78	31.07	—	—	—
Claude Sonnet 4.52026.02		0.95	81.87	93.19	82.2	4.35	81.81	—	—	—
Gemma-2-Instruct2026.03		0.946	—	60	—	—	—	0.991	1.048	0.913
OpenAI GPT 5.22026.02		0.94	84.66	85.24	92.59	12.25	53.95	—	—	—
Qwen2.5 7B ESMA2026.02		0.94	69.86	60.71	67.64	27.4	35.87	—	—	—
Qwen2.5 1.5B ESMA2026.02		0.93	68.86	41.86	37.89	35.86	28.26	—	—	—
Gemma3 4B ESMA2026.02		0.92	67.94	55.58	53.29	27.92	36.77	—	—	—
Llama3.2 3B ESMA2026.02		0.89	67.73	55.64	58.1	29.89	35.58	—	—	—
Gemini 3 Flash2026.02		0.68	90.71	92.68	96.34	6.72	76.98	—	—	—
Qwen2.5 7B2026.02		0.64	61.65	50.43	36.2	33.31	41.21	—	—	—
Qwen2.5 3B2026.02		0.29	62.7	35.67	18.22	54.46	33.47	—	—	—
Qwen2.5 1.5B2026.02		0.2	53.3	42.86	53.81	53.56	38.69	—	—	—
Llama3.2 3B2026.02		0.2	49.57	53.74	14.6	38.66	52.44	—	—	—
Gemma3 4B2026.02		0.04	52.75	46.53	19.12	51.88	46.15	—	—	—