Long context

AA-LCR leaderboard

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

As of 2026-10-07, the highest measured score on AA-LCR is 88.7% by Kimi K3. 8 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Kimi K3	88.7%	measured
2	Step 5 Preview	88.3%	measured
3	MiMo-V2.6-Pro	86.3%	measured
4	Claude Fable 5.1	85.3%	measured
5	Claude Opus 5.5	84.7%	measured
6	GPT-5.5	84.3%	measured
7	DeepSeek V4.1 Flash	84.0%	measured
8	GPT-5.6 Sol	84.0%	measured
9	GPT-5.6 Luna	83.7%	measured
10	GPT-6 Sol	83.7%	measured
11	GPT-5.3 Codex	83.3%	measured
12	GPT-6 Luna	83.3%	measured
13	Muse Glimmer 30B	83.3%	measured
14	GPT-5.6 Terra	83.0%	measured
15	GPT-6.1 Sol	83.0%	measured
16	Ling 3.1 Flash	83.0%	measured
17	MiniMax M3	83.0%	measured
18	Muse Spark 1.3	83.0%	measured
19	Claude Haiku 5.5	82.7%	measured
20	Claude Sonnet 5.5	82.7%	measured
21	GPT-5.2	82.7%	measured
22	Claude Fable 5	82.3%	measured
23	GPT-5.2-Codex	82.3%	measured
24	Claude Sonnet 5	82.0%	measured
25	Gemini 3.1 Pro	82.0%	measured
26	GPT-5.4	82.0%	measured
27	Qwen3.8-27B	82.0%	measured
28	Gemini 3.7 Flash	81.7%	measured
29	Gemini 3.8 Flash	81.3%	measured
30	Mistral Large 4	81.3%	measured
31	Kimi K2.6	81.0%	measured
32	GPT-6 Astra	80.7%	measured
33	Qwen 3.6 Max (preview)	80.7%	measured
34	DeepSeek V4 Pro 0813	80.3%	measured
35	Grok 4.6	80.3%	measured
36	Qwen3.8 Max Preview	80.3%	measured
37	Gemini 3.6 Flash	80.0%	measured
38	GLM-5.3-Flash	80.0%	measured
39	GPT-5.1	80.0%	measured
40	DeepSeek V4 Flash 0731	79.7%	measured
41	Gemini 4 Argon	79.7%	measured
42	GLM-5.3	79.7%	measured
43	MiMo-V2.5-Pro	79.7%	measured
44	Qwen3.8-Flash-Next	79.7%	measured
45	Apodex 1.1	79.3%	measured
46	Apodex 1.1 Mini	79.3%	measured
47	Claude Opus 5	79.3%	measured
48	Grok 4.5	79.3%	measured
49	Kimi K2.7 Code	79.3%	measured
50	Hy3	79.0%	measured
51	Muse Spark 1.2	79.0%	measured
52	Qwen3.7 Max	79.0%	measured
53	Claude Opus 4.7 (Adaptive)	78.7%	measured
54	GLM-5.2	78.3%	measured
55	Ling 3.0 Flash VL	78.3%	measured
56	MiniMax M2.7	78.3%	measured
57	Qwen3.6 Plus	78.3%	measured
58	GPT-5 (high)	78.2%	measured
59	Claude Opus 4.6 (Adaptive)	78.0%	measured
60	Kimi K2.5	78.0%	measured
61	Kimi K2.5 (Reasoning)	78.0%	measured
62	Muse Spark	78.0%	measured
63	Claude Opus 4.8	77.7%	measured
64	Muse Spark 1.1	77.7%	measured
65	Qwen3.5-27B	77.7%	measured
66	Claude Opus 4.5 Thinking	77.3%	measured
67	Inkling	77.3%	measured
68	Qwen3.6-27B	77.3%	measured
69	GPT-5.4 mini	77.0%	measured
70	Ternary Bonsai 2 27B	77.0%	measured
71	GPT-5.4 nano	76.7%	measured
72	Grok 4.7	76.7%	measured
73	Quasar 438B	76.3%	measured
74	Qwen3.5-122B-A10B	76.3%	measured
75	Claude 4.1 Opus Thinking	76.0%	measured
76	Gemini 3.5 Flash-Lite	76.0%	measured
77	Gemini 3 Pro	76.0%	measured
78	GPT-5 (medium)	76.0%	measured
79	Sakana Fugu-Ultra	75.9%	estimated ± 4.5 pp, low confidence
80	Qwen3.8 Max	75.8%	estimated ± 4.5 pp, low confidence
81	Claude Opus 4.7	75.7%	measured
82	GLM-5	75.7%	measured
83	Inkling-Small	75.7%	measured
84	MiMo-V2-Omni	75.0%	measured
85	Sakana Fugu	74.9%	estimated ± 4.5 pp, medium confidence
86	o3	74.7%	measured
87	MiMo-V2.6-Flash	74.3%	measured
88	Grok 4.1 Fast (Reasoning)	74.0%	measured
89	GLM-5.1	73.7%	measured
90	Grok 4 Fast (Reasoning)	73.7%	measured
91	Step 3.7 Flash	73.7%	measured
92	Agents-A1	73.3%	estimated ± 7.0 pp, low confidence
93	Beam	73.3%	estimated ± 7.0 pp, low confidence
94	Agents-A1-4B	73.3%	estimated ± 7.0 pp, low confidence
95	Ling 3.0 Flash	73.0%	measured
96	Ling 3.0 Flash FP8	73.0%	measured
97	Qwen3.7 Plus	73.0%	measured
98	Qwen3.5-35B-A3B	72.0%	measured
99	GLM-5-Turbo	71.7%	measured
100	Qwen3.6-35B-A3B	71.7%	measured
101	GLM-4.7	71.0%	measured
102	Solar Pro 4	71.0%	measured
103	Claude Opus 4.5	70.7%	measured
104	GLM-5V-Turbo	70.3%	measured
105	Pokee-Isaac 28B	69.8%	estimated ± 4.5 pp, medium confidence
106	Gemma 4 31B	69.7%	measured
107	Gemini 3.5 Flash	69.3%	measured
108	GPT-5.1-Codex	69.3%	measured
109	GPT-5.1-Codex-Max	69.3%	measured
110	Mistral Medium 3.5 128B	69.3%	measured
111	Gemini 2.5 Pro	69.0%	measured
112	Claude Sonnet 4.6	68.3%	measured
113	GPT-4.1	68.3%	measured
114	MiMo-V2-Pro	68.3%	measured
115	Grok 4	68.0%	measured
116	Mercury 2.5	68.0%	measured
117	Claude Opus 4.6	67.0%	measured
118	Nemotron 3 Ultra	67.0%	measured
119	Hy3 Preview	66.7%	measured
120	A.X K2	66.0%	measured
121	Gemma 4 26B A4B	65.7%	measured
122	Nemotron 3 Super 100B	65.7%	measured
123	o1	65.0%	measured
124	Grok 4.3	64.3%	measured
125	Qwen3.5 397B	64.3%	measured
126	Qwen3.5 397B (Reasoning)	64.3%	measured
127	Gemma 4 12B	63.7%	measured
128	Solar Open 2	62.3%	measured
129	K-Exaone	61.3%	measured
130	Ling 3.0 Tiny	60.3%	measured
131	MiniCPM5-2B	59.0%	measured
132	DeepSeek V3.1 (Reasoning)	56.7%	measured
133	LLaDA2.2-mini	56.5%	estimated ± 7.0 pp, low confidence
134	K-EXAONE 2.0	56.2%	measured
135	DeepSeek-R1	55.7%	measured
136	Gemini 3 Flash	55.3%	measured
137	Grok Code Fast 1	53.0%	measured
138	Kimi K2	53.0%	measured
139	Command A+	52.7%	measured
140	GPT-OSS 120B	52.0%	measured
141	Llama 4 Maverick	50.0%	measured
142	Qwen3 Max	50.0%	measured
143	Gemini 2.5 Flash	49.9%	measured
144	Mistral Small 4	49.7%	measured
145	Mistral Small 4 (Reasoning)	49.7%	measured
146	GPT-4o	49.3%	measured
147	Nemotron 3.5 Lightning 30B A3B NVFP4	49.2%	measured
148	Granite 4.2 30B	49.0%	measured
149	DeepSeek V3.1	47.0%	measured
150	GLM-4.5-Air	46.7%	measured
151	DeepSeek V3.2	45.7%	measured
152	Granite 4.2 8B	45.0%	measured
153	Claude 4 Sonnet	44.0%	measured
154	GPT-4.1 mini	44.0%	measured
155	DeepSeek V3 0324	40.7%	measured
156	Nemotron 3 Nano Omni 30B A3B	39.7%	measured
157	Celeris-1	38.0%	measured
158	Nemotron 3 Nano 30B	38.0%	measured
159	Trinity-Large-Preview	38.0%	measured
160	Trinity-Large-Thinking	38.0%	measured
161	North Mini Code	37.3%	measured
162	MiMo-V2-Flash	36.3%	measured
163	Mistral Large 3	36.0%	measured
164	GPT-OSS 20B	34.7%	measured
165	Solar Pro 3	32.3%	measured
166	Gemma 4 E4B	32.0%	measured
167	Grok 4.1 Fast	31.3%	measured
168	Ling 2.6 Flash	31.3%	measured
169	Mistral Medium 3	31.3%	measured
170	DeepSeek V3	29.3%	measured
171	Claude 3 Haiku	27.7%	measured
172	Llama 4 Scout	27.7%	measured
173	GLM-4.6	26.3%	measured
174	Llama 3.1 405B	25.3%	measured
175	Granite 4.2 3B	24.3%	measured
176	Nova Pro	21.0%	measured
177	GPT-4.1 nano	20.3%	measured
178	Gemma 4 E2B	16.3%	measured
179	Ultravox v0.6 Llama 3.3 70B	15.7%	measured
180	DeepSeek R1 Distill Qwen 32B	8.7%	measured
181	Gemma 3 27B	7.3%	measured
182	Granite-4.0-H-1B	7.0%	measured
183	LFM2.5-2.6B	5.7%	measured
184	Exaone 4.0 1.2B	0.0%	measured
185	Granite-4.0-350M	0.0%	measured
186	Granite-4.0-H-350M	0.0%	measured
187	LFM2.5-8B-A1B	0.0%	measured
188	LFM2.5-VL-1.6B-Extract	0.0%	measured
189	Phi-4	0.0%	measured
190	Qwen3-Omni-30B-A3B-Instruct	0.0%	measured
191	Qwen3-Omni-30B-A3B-Thinking	0.0%	measured
192	Sarvam 105B	0.0%	measured
193	Sarvam 30B	0.0%	measured
194	Solar Pro 2	0.0%	measured

AA-LCR leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General