Knowledge & reasoning

AA-HLE leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on AA-HLE is 61.4% by Claude Opus 5.5. 47 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 5.5	61.4%	measured
2	Claude Fable 5.1	59.1%	measured
3	Gemini 4 Argon	57.1%	measured
4	Claude Fable 5	55.5%	measured
5	Claude Mythos 5	55.2%	estimated ± 5.0 pp, high confidence
6	Claude Sonnet 5.5	55.0%	measured
7	Claude Opus 5	54.9%	measured
8	GPT-6 Astra	54.7%	measured
9	GPT-6.1 Sol	52.9%	measured
10	Sakana Fugu	52.9%	estimated ± 4.0 pp, high confidence
11	Sakana Fugu-Ultra	52.9%	estimated ± 4.0 pp, high confidence
12	GPT-5.5 Pro	51.3%	estimated ± 4.6 pp, high confidence
13	GPT-5.4 Pro	51.0%	estimated ± 4.6 pp, high confidence
14	GPT-5.6 Sol	49.5%	measured
15	MiMo-V2.6-Pro	49.4%	measured
16	Gemini 3 Pro Deep Think	48.8%	estimated ± 4.6 pp, high confidence
17	Claude Opus 4.8	48.7%	measured
18	Muse Spark 1.3	48.7%	measured
19	Gemini 3.7 Flash	47.9%	measured
20	GPT-6 Sol	47.9%	measured
21	Gemini 3.8 Flash	47.8%	measured
22	dots3-note Preview	47.3%	estimated ± 6.3 pp, medium confidence
23	Pareto 26.9	47.1%	estimated ± 5.0 pp, high confidence
24	Gemini 3.1 Pro	47.0%	measured
25	Kimi K3	46.9%	measured
26	Step 5 Preview	46.5%	measured
27	Muse Spark 1.1	46.2%	measured
28	GPT-5.5	45.8%	measured
29	Muse Spark 1.2	45.5%	measured
30	Claude Haiku 5.5	44.4%	measured
31	GPT-5.4	43.7%	measured
32	Grok 4.7	43.1%	measured
33	Qwen3.8 Max Preview	43.1%	measured
34	Ornith-1.5-397B	43.1%	estimated ± 4.0 pp, high confidence
35	GPT-5.6 Terra	42.9%	measured
36	Grok 4.6	42.9%	measured
37	Gemini 3.5 Flash	42.7%	measured
38	Grok 4.5	42.7%	measured
39	GPT-5.3 Codex	42.5%	measured
40	Qwen3.8 Max	42.5%	estimated ± 4.0 pp, high confidence
41	Claude Opus 4.7 (Adaptive)	42.3%	measured
42	GLM-5.3	42.3%	measured
43	Pareto 26.10 Preview	41.8%	estimated ± 4.0 pp, high confidence
44	Hy4 preview	41.5%	estimated ± 4.0 pp, high confidence
45	Claude Sonnet 5	41.3%	measured
46	GLM-5.2	41.1%	measured
47	DeepSeek V4 Pro 0813	41.0%	measured
48	Gemini 3.6 Flash	40.8%	measured
49	Muse Spark	40.7%	measured
50	Grok 4.20	40.7%	estimated ± 2.6 pp, medium confidence
51	Qwen3.7 Max	40.5%	measured
52	Claude Opus 4.6 (Adaptive)	39.9%	measured
53	GLM-5.3-Flash	39.9%	measured
54	Gemini 3 Pro	39.7%	measured
55	GPT-5.6 Luna	39.5%	measured
56	Ling 3.1 Flash	39.4%	measured
57	DeepSeek V4.1 Flash	39.2%	measured
58	MiniMax M3	39.0%	measured
59	DeepSeek V4 Flash 0731	38.6%	measured
60	GPT-6 Luna	38.5%	measured
61	Qwen3.8-Flash-Next	38.0%	measured
62	Qwen3.8-Omni-Flash	37.9%	estimated ± 4.0 pp, high confidence
63	GPT-5.2	37.7%	measured
64	Agents-A1	37.7%	estimated ± 8.4 pp, medium confidence
65	Kimi K2.6	37.5%	measured
66	Grok 4.3	37.2%	measured
67	Beam	36.6%	estimated ± 4.0 pp, high confidence
68	GPT-5.2-Codex	35.7%	measured
69	MiMo-V2.5-Pro	35.7%	measured
70	Qwen3.7 Plus	35.6%	measured
71	Interfaze Beta	35.1%	estimated ± 4.0 pp, high confidence
72	MiMo-V2.6-Flash	35.1%	measured
73	Kimi K2.7 Code	35.0%	measured
74	Mistral Large 4	35.0%	measured
75	Apodex 1.1	34.1%	measured
76	Apodex 1.1 Mini	34.1%	measured
77	Qwen3.8-27B	33.9%	measured
78	Ornith-1.5-35B-A3B	33.5%	estimated ± 4.0 pp, high confidence
79	Hy3	33.5%	measured
80	Hy3 Preview	33.5%	measured
81	Claude Opus 4.7	33.3%	measured
82	Inkling-Small	33.3%	measured
83	Inkling	31.9%	measured
84	Claude Sonnet 4.5	30.9%	estimated ± 6.3 pp, medium confidence
85	Qwen 3.6 Max (preview)	30.8%	measured
86	Kimi K2.5	30.7%	measured
87	Kimi K2.5 (Reasoning)	30.7%	measured
88	MiMo-V2-Pro	30.4%	measured
89	Claude Opus 4.5 Thinking	30.1%	measured
90	GLM-5.1	30.1%	measured
91	Claude Sonnet 4.5 Thinking	29.7%	estimated ± 6.2 pp, medium confidence
92	A.X K2	29.6%	measured
93	MiniMax M2.7	29.6%	measured
94	GLM-5	29.3%	measured
95	Solar Pro 4	29.2%	measured
96	GPT-5.1	28.5%	measured
97	GPT-5 (high)	28.5%	measured
98	Nemotron 3 Ultra	28.4%	measured
99	GPT-5.4 nano	28.3%	measured
100	GPT-5.4 mini	28.1%	measured
101	Ornith-1.5-9B	27.9%	estimated ± 4.0 pp, high confidence
102	GLM-5-Turbo	27.8%	measured
103	Qwen3.6 Plus	27.8%	measured
104	Solar Open 2	27.7%	estimated ± 4.0 pp, high confidence
105	GLM-4.7	27.4%	measured
106	Ternary Bonsai 2 27B	26.8%	estimated ± 4.0 pp, high confidence
107	Grok 4	26.7%	measured
108	GPT-5.1-Codex	25.7%	measured
109	GPT-5.1-Codex-Max	25.7%	measured
110	Qwen3.5 Flash	25.7%	estimated ± 7.4 pp, medium confidence
111	GPT-5 (medium)	25.4%	measured
112	Qwen3.5-122B-A10B	25.2%	measured
113	MAI-Thinking-1	24.3%	estimated ± 4.0 pp, high confidence
114	MiMo-V2.5	24.0%	estimated ± 7.4 pp, medium confidence
115	Qwen3.5-27B	23.9%	measured
116	o3-pro	23.8%	estimated ± 4.0 pp, high confidence
117	Ling 3.0 Flash	23.7%	measured
118	Ling 3.0 Flash FP8	23.7%	measured
119	Gemma 4 31B	23.6%	measured
120	Gemini 3.1 Flash-Lite	23.4%	estimated ± 7.4 pp, medium confidence
121	Qwen3.6-27B	23.1%	measured
122	Gemini 2.5 Pro	22.5%	measured
123	Qwen3.6-35B-A3B	22.2%	measured
124	MiMo-V2-Omni	22.1%	measured
125	Ling 3.0 Flash VL	22.0%	measured
126	Muse Glimmer 30B	22.0%	measured
127	Step 3.7 Flash	21.4%	measured
128	Qwen3.5-35B-A3B	21.0%	measured
129	Nemotron 3 Super 100B	20.8%	measured
130	o3	20.1%	measured
131	Qwen3.5 397B	19.8%	measured
132	Qwen3.5 397B (Reasoning)	19.8%	measured
133	GPT-OSS 120B	19.6%	measured
134	Gemma 4 26B A4B	19.3%	measured
135	Grok 4.1 Fast (Reasoning)	19.3%	measured
136	Claude Opus 4.6	19.1%	measured
137	Grok 4 Fast (Reasoning)	19.1%	measured
138	Gemini 3.5 Flash-Lite	18.8%	measured
139	Claude 4.1 Opus	18.7%	estimated ± 4.4 pp, high confidence
140	Quasar 438B	18.7%	measured
141	K-EXAONE 2.0	18.6%	measured
142	GLM-5V-Turbo	17.1%	measured
143	DeepSeek-R1	15.8%	measured
144	Trinity-Large-Preview	15.8%	measured
145	Trinity-Large-Thinking	15.8%	measured
146	Gemma 4 12B	15.7%	measured
147	Claude Haiku 4.5	15.6%	estimated ± 7.4 pp, medium confidence
148	GLM-4.5	15.6%	estimated ± 7.4 pp, medium confidence
149	Gemini 3 Flash	15.0%	measured
150	DeepSeek V3.1 (Reasoning)	14.3%	measured
151	Qwen3 235B 2507	14.3%	estimated ± 6.5 pp, medium confidence
152	K-Exaone	13.9%	measured
153	Mistral Medium 3.5 128B	13.8%	measured
154	Claude Sonnet 4.6	13.3%	measured
155	Claude Opus 4.5	13.2%	measured
156	Claude 4.1 Opus Thinking	12.5%	measured
157	o1-preview	12.2%	estimated ± 4.0 pp, high confidence
158	Command A+	12.0%	measured
159	Qwen3 Max	11.9%	measured
160	Mercury 2.5	11.8%	measured
161	Nemotron 3 Nano 30B	11.4%	measured
162	Laguna XS.2	11.4%	estimated ± 7.4 pp, medium confidence
163	DeepSeek V3.2	11.2%	measured
164	Granite 4.2 30B	11.2%	measured
165	North Mini Code	11.1%	measured
166	GPT-OSS 20B	11.0%	measured
167	Sarvam 105B	11.0%	measured
168	Laguna M.1	10.7%	estimated ± 7.4 pp, low confidence
169	Nemotron 3.5 Lightning 30B A3B NVFP4	10.6%	measured
170	ZAYA1-8B	10.5%	estimated ± 4.0 pp, high confidence
171	Solar Pro 3	10.3%	measured
172	o1-pro	10.2%	estimated ± 4.4 pp, high confidence
173	Mistral Small 4	9.9%	measured
174	Mistral Small 4 (Reasoning)	9.9%	measured
175	Granite 4.2 8B	9.7%	measured
176	LongCat-Flash-Lite-Sparse	9.5%	estimated ± 4.0 pp, high confidence
177	Ling 3.0 Tiny	9.3%	measured
178	MiniCPM5-2B	8.9%	measured
179	MiMo-V2-Flash	8.6%	measured
180	Grok Code Fast 1	8.0%	measured
181	o3-mini	7.9%	measured
182	Qwen3-Omni-30B-A3B-Thinking	7.5%	measured
183	Sarvam 30B	7.5%	measured
184	Kimi K2	7.4%	measured
185	Nemotron Ultra 253B	7.4%	measured
186	GLM-4.5-Air	7.0%	measured
187	o1	7.0%	measured
188	LFM2.5-8B-A1B	6.9%	measured
189	Celeris-1	6.8%	measured
190	DeepSeek V3.1	6.7%	measured
191	Granite 4.2 3B	6.6%	measured
192	Granite-4.0-H-350M	6.4%	measured
193	Ling 2.6 Flash	6.3%	measured
194	LFM2.5-2.6B	6.2%	measured
195	Exaone 4.0 1.2B	5.7%	measured
196	GLM-4.6	5.5%	measured
197	Granite-4.0-350M	5.5%	measured
198	Grok 4.1 Fast	5.1%	measured
199	LFM2.5-VL-1.6B-Extract	5.1%	measured
200	Exaone 4.0 32B	5.0%	measured
201	GPT-4.1 mini	5.0%	measured
202	Granite-4.0-H-1B	5.0%	measured
203	Phi-4 Multimodal Instruct	5.0%	measured
204	Claude 3.5 Sonnet	4.9%	estimated ± 6.5 pp, medium confidence
205	Llama 4 Maverick	4.9%	measured
206	Gemma 4 E2B	4.8%	measured
207	Nemotron 3 Nano Omni 30B A3B	4.8%	measured
208	DeepSeek V3 0324	4.7%	measured
209	Gemini 2.5 Flash	4.7%	measured
210	DeepSeek R1 Distill Qwen 32B	4.6%	measured
211	Gemini 1.5 Pro	4.6%	measured
212	Qwen3-Omni-30B-A3B-Instruct	4.6%	measured
213	Gemma 3 27B	4.4%	measured
214	Claude 4 Sonnet	4.3%	measured
215	Gemini 1.0 Pro	4.2%	measured
216	GPT-4.1	4.2%	measured
217	GPT-4o mini	4.2%	measured
218	Mistral Large 3	4.2%	measured
219	Claude 3 Haiku	4.1%	measured
220	Mistral Medium 3	4.1%	measured
221	Llama 3.1 405B	4.0%	measured
222	Gemma 4 E4B	3.8%	measured
223	GPT-4.1 nano	3.8%	measured
224	Llama 4 Scout	3.8%	measured
225	Phi-4	3.8%	measured
226	Solar Pro 2	3.7%	measured
227	Ultravox v0.6 Llama 3.3 70B	3.6%	measured
228	Qwen2.5 Coder 32B Instruct	3.5%	measured
229	Mistral Large 2	3.3%	measured
230	Mellum2-12B-A2.5B-Thinking	3.3%	estimated ± 4.0 pp, medium confidence
231	Nova Pro	3.2%	measured
232	ZAYA1-74B-Preview	3.2%	estimated ± 4.0 pp, medium confidence
233	GPT-4 Turbo	3.1%	measured
234	DeepSeek V3	2.9%	measured
235	Claude 3 Opus	2.8%	measured
236	GPT-4o	2.4%	measured
237	LFM2.5-230M	0.0%	estimated ± 4.0 pp, medium confidence
238	LFM2.5-VL-450M	0.0%	estimated ± 6.5 pp, low confidence
239	LLaDA2.2-mini	0.0%	estimated ± 4.0 pp, medium confidence
240	Mellum2-12B-A2.5B-Instruct	0.0%	estimated ± 4.0 pp, medium confidence
241	MiniCPM5-1B	0.0%	estimated ± 4.0 pp, medium confidence
242	Soofi S 30B-A3B	0.0%	estimated ± 4.0 pp, medium confidence

AA-HLE leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General