Knowledge Probing on Probe 1

96AccuracyAnthropic.Claude-3-Haiku

Evaluation results

Method
Anthropic.Claude-3-Haiku2026.01	96	96	91	91	91
Llama-3.3-70B-Instruct2026.01	95	95	96	96	96
Anthropic.Claude-3-Haiku2026.01	93	93	94	93	93
Anthropic.Claude-3-Haiku2026.01	93	93	94	94	93
Anthropic.Claude-3-Haiku2026.01	92	92	93	93	93
Llama-3.3-70B-Instruct2026.01	92	92	93	92	92
Llama-3.3-70B-Instruct2026.01	92	92	92	92	92
GPT-OSS-20B2026.01	91	91	92	92	91
Llama-3.3-70B-Instruct2026.01	91	91	91	91	91
Llama-3.3-70B-Instruct2026.01	91	91	91	90	90
Qwen2.5-32B-Instruct2026.01	91	91	92	90	91
Qwen2.5-32B-Instruct2026.01	90	90	90	89	89
Anthropic.Claude-3-Haiku2026.01	89	89	91	90	90
Qwen2.5-32B-Instruct2026.01	89	89	89	88	88
Qwen2.5-32B-Instruct2026.01	89	89	89	87	88
Mixtral-8x7B-v0.12026.01	88	88	90	89	89
GPT-OSS-20B2026.01	87	87	88	87	87
Mixtral-8x7B-v0.12026.01	87	87	88	87	87
Qwen2.5-32B-Instruct2026.01	87	87	87	87	87
Llama-3.3-70B-Instruct2026.01	86	86	72	71	71
GPT-OSS-20B2026.01	85	85	87	86	86
Qwen2.5-32B-Instruct2026.01	85	85	84	84	84
Anthropic.Claude-3-Haiku2026.01	84	84	87	84	84
GPT-OSS-20B2026.01	84	84	87	85	85
Mixtral-8x7B-v0.12026.01	84	84	86	85	84
Llama-3.1-8B-Instruct2026.01	83	83	85	82	83
Mixtral-8x7B-v0.12026.01	83	83	86	83	84
Mixtral-8x7B-v0.12026.01	83	83	82	80	80
GPT-OSS-20B2026.01	82	82	85	83	82
GPT-OSS-20B2026.01	82	82	84	82	82
Mixtral-8x7B-v0.12026.01	79	79	85	79	80
Mistral-7B-Instruct-v0.32026.01	73	73	78	73	74
Mistral-7B-Instruct-v0.32026.01	72	72	78	72	73
Llama-3.1-8B-Instruct2026.01	69	69	76	66	67
Mistral-7B-Instruct-v0.32026.01	69	69	77	70	71
Mistral-7B-Instruct-v0.32026.01	68	68	75	69	70
Llama-3.1-8B-Instruct2026.01	67	67	74	65	66
Mistral-7B-Instruct-v0.32026.01	67	67	79	67	70
Mistral-7B-Instruct-v0.32026.01	67	67	78	66	69
Llama-3.1-8B-Instruct2026.01	66	66	71	65	65
Llama-3.1-8B-Instruct2026.01	65	65	74	63	65
Llama-3.1-8B-Instruct2026.01	63	63	69	62	62