Agentic · tools

MazeBench leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on MazeBench is 23.0% by GPT-6 Astra. 134 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 4 Carbon	25.5%	estimated ± 3.0 pp, medium confidence
2	GPT-6 Astra	23.0%	measured
3	GPT-6.1 Sol	17.0%	measured
4	Muse Wonder Full	17.0%	measured
5	GPT-6.1 Terra	16.0%	measured
6	Claude Opus 5.5	13.0%	measured
7	GPT-5.6 Sol	13.0%	measured
8	Claude Opus 5	12.0%	measured
9	Gemini 4 Argon	12.0%	measured
10	Claude Fable 5	11.0%	measured
11	Claude Fable 5.1	10.0%	measured
12	GPT-5.5 Pro	8.1%	estimated ± 6.1 pp, low confidence
13	Claude Sonnet 5.5	8.0%	measured
14	Qwen3.8 Max Preview	7.7%	estimated ± 5.4 pp, low confidence
15	Ling 3.1 Flash	7.7%	estimated ± 5.4 pp, low confidence
16	GPT-5.4 Pro	7.0%	estimated ± 6.1 pp, low confidence
17	GPT-6 Sol	7.0%	measured
18	Ling 3.0 Flash Fin	6.3%	estimated ± 6.6 pp, low confidence
19	GPT-5.6 Terra	6.0%	measured
20	Claude Mythos 5	5.2%	estimated ± 6.1 pp, low confidence
21	Gemini 3.8 Flash	4.0%	measured
22	MiMo-V2.6-Pro	3.4%	estimated ± 3.0 pp, high confidence
23	Claude Mythos Preview	3.2%	estimated ± 3.0 pp, high confidence
24	DeepSeek V4 Pro 0813	2.7%	estimated ± 4.0 pp, high confidence
25	Hy4 preview	2.7%	estimated ± 4.0 pp, high confidence
26	Step 5 Preview	2.7%	estimated ± 4.0 pp, high confidence
27	Kimi K3.1 Pro	2.0%	measured
28	Qwen3.8-Flash-Next	2.0%	estimated ± 4.0 pp, high confidence
29	Claude Opus 4.7	1.7%	estimated ± 4.9 pp, high confidence
30	Claude Opus 4.6	1.2%	estimated ± 4.9 pp, high confidence
31	Claude Haiku 5.5	1.0%	measured
32	DeepSeek V4.1 Flash 1017	1.0%	measured
33	Gemini 3.1 Pro	1.0%	measured
34	Gemini 3.7 Flash	1.0%	measured
35	Grok 4.6	1.0%	measured
36	Grok 4.7	1.0%	measured
37	Kimi K3	1.0%	measured
38	MiniMax M3.1 Flash	1.0%	estimated ± 5.4 pp, low confidence
39	Mistral Large 4	0.5%	estimated ± 5.4 pp, low confidence
40	Grok 4.5	0.4%	estimated ± 5.4 pp, low confidence
41	Hy3 Preview	0.4%	estimated ± 5.4 pp, low confidence
42	Apodex 1.1	0.4%	estimated ± 5.4 pp, low confidence
43	Apodex 1.1 Mini	0.4%	estimated ± 5.4 pp, low confidence
44	Quasar 438B	0.4%	estimated ± 5.4 pp, low confidence
45	Beam 1.0 501B-A23B	0.4%	estimated ± 5.4 pp, low confidence
46	Nemotron 3.5 Super VL 120B A12B	0.4%	estimated ± 5.4 pp, low confidence
47	Nemotron 3.5 Super VL 120B A12B NVFP4	0.4%	estimated ± 5.4 pp, low confidence
48	Ling 3.0 Flash VL	0.4%	estimated ± 5.4 pp, low confidence
49	Qwen3.7 Max	0.4%	estimated ± 5.4 pp, low confidence
50	Inkling-Small	0.4%	estimated ± 5.4 pp, low confidence
51	GLM-5.1	0.4%	estimated ± 5.4 pp, low confidence
52	Solar Pro 4	0.4%	estimated ± 5.4 pp, low confidence
53	Grok 4.3	0.4%	estimated ± 5.4 pp, low confidence
54	Inkling	0.4%	estimated ± 5.4 pp, low confidence
55	Hy3	0.4%	estimated ± 5.4 pp, low confidence
56	Kimi K2.6	0.4%	estimated ± 5.4 pp, low confidence
57	Kimi K2.7 Code	0.4%	estimated ± 5.4 pp, low confidence
58	Step 3.7 Flash	0.4%	estimated ± 5.4 pp, low confidence
59	GLM-4.7	0.4%	estimated ± 5.4 pp, low confidence
60	GPT-5.4 mini	0.4%	estimated ± 5.4 pp, low confidence
61	MiniMax M2.7	0.4%	estimated ± 5.4 pp, low confidence
62	Muse Spark	0.4%	estimated ± 5.4 pp, low confidence
63	Qwen3.6 Plus	0.4%	estimated ± 5.4 pp, low confidence
64	Qwen3.6-27B	0.4%	estimated ± 5.4 pp, low confidence
65	Gemini 3.5 Flash-Lite	0.4%	estimated ± 5.4 pp, low confidence
66	Nemotron 3.5 Bolt 74B-A9B	0.4%	estimated ± 5.4 pp, low confidence
67	A.X K2	0.4%	estimated ± 5.4 pp, low confidence
68	GPT-5.4 nano	0.4%	estimated ± 5.4 pp, low confidence
69	Qwen3.6-35B-A3B	0.4%	estimated ± 5.4 pp, low confidence
70	Kimi K2.5 (Reasoning)	0.4%	estimated ± 5.4 pp, low confidence
71	GPT-5.1	0.4%	estimated ± 5.4 pp, low confidence
72	Qwen3.5-122B-A10B	0.4%	estimated ± 5.4 pp, low confidence
73	Celeris-1	0.4%	estimated ± 5.4 pp, low confidence
74	Claude 3.5 Haiku	0.4%	estimated ± 5.4 pp, low confidence
75	Command A+	0.4%	estimated ± 5.4 pp, low confidence
76	DeepSeek V3	0.4%	estimated ± 5.4 pp, low confidence
77	DeepSeek V3 0324	0.4%	estimated ± 5.4 pp, low confidence
78	Gemini 2.5 Pro	0.4%	estimated ± 5.4 pp, low confidence
79	Gemma 3 27B	0.4%	estimated ± 5.4 pp, low confidence
80	Gemma 4 12B	0.4%	estimated ± 5.4 pp, low confidence
81	Gemma 4 26B A4B	0.4%	estimated ± 5.4 pp, low confidence
82	Gemma 4 31B	0.4%	estimated ± 5.4 pp, low confidence
83	Gemma 4 E2B	0.4%	estimated ± 5.4 pp, low confidence
84	Gemma 4 E4B	0.4%	estimated ± 5.4 pp, low confidence
85	GPT-4.1 mini	0.4%	estimated ± 5.4 pp, low confidence
86	GPT-4.1 nano	0.4%	estimated ± 5.4 pp, low confidence
87	GPT-4o	0.4%	estimated ± 5.4 pp, low confidence
88	GPT-4o mini	0.4%	estimated ± 5.4 pp, low confidence
89	GPT-OSS 120B	0.4%	estimated ± 5.4 pp, low confidence
90	GPT-OSS 20B	0.4%	estimated ± 5.4 pp, low confidence
91	Granite 4.2 30B	0.4%	estimated ± 5.4 pp, low confidence
92	Granite 4.2 3B	0.4%	estimated ± 5.4 pp, low confidence
93	Granite 4.2 8B	0.4%	estimated ± 5.4 pp, low confidence
94	K-Exaone	0.4%	estimated ± 5.4 pp, low confidence
95	K-EXAONE 2.0	0.4%	estimated ± 5.4 pp, low confidence
96	LFM2.5-2.6B	0.4%	estimated ± 5.4 pp, low confidence
97	Ling 2.6 Flash	0.4%	estimated ± 5.4 pp, low confidence
98	Ling 3.0 Tiny	0.4%	estimated ± 5.4 pp, low confidence
99	Llama 3.3 70B	0.4%	estimated ± 5.4 pp, low confidence
100	Llama 4 Scout	0.4%	estimated ± 5.4 pp, low confidence
101	Mercury 2.5	0.4%	estimated ± 5.4 pp, low confidence
102	MiMo-V2-Flash	0.4%	estimated ± 5.4 pp, low confidence
103	MiniCPM5-2B	0.4%	estimated ± 5.4 pp, low confidence
104	Mistral Large 3	0.4%	estimated ± 5.4 pp, low confidence
105	Mistral Medium 3.5 128B	0.4%	estimated ± 5.4 pp, low confidence
106	Mistral Small 3.1	0.4%	estimated ± 5.4 pp, low confidence
107	Mistral Small 4	0.4%	estimated ± 5.4 pp, low confidence
108	Mistral Small 4 (Reasoning)	0.4%	estimated ± 5.4 pp, low confidence
109	Muse Glimmer 30B	0.4%	estimated ± 5.4 pp, low confidence
110	Nemotron 3.5 Lightning 30B A3B NVFP4	0.4%	estimated ± 5.4 pp, low confidence
111	Nemotron 3 Nano 30B	0.4%	estimated ± 5.4 pp, low confidence
112	Nemotron 3 Nano Omni 30B A3B	0.4%	estimated ± 5.4 pp, low confidence
113	North Mini Code	0.4%	estimated ± 5.4 pp, low confidence
114	Qwen3.7 Plus	0.4%	estimated ± 5.4 pp, low confidence
115	Solar Pro 3	0.4%	estimated ± 5.4 pp, low confidence
116	Trinity-Large-Preview	0.4%	estimated ± 5.4 pp, low confidence
117	Trinity-Large-Thinking	0.4%	estimated ± 5.4 pp, low confidence
118	Ultravox v0.6 Llama 3.3 70B	0.4%	estimated ± 5.4 pp, low confidence
119	Ultramarine 1 Flash	0.0%	estimated ± 5.1 pp, medium confidence
120	Atria Dawn Preview	0.0%	estimated ± 5.1 pp, medium confidence
121	Claude Opus 4.7 (Adaptive)	0.0%	estimated ± 5.1 pp, medium confidence
122	GPT-5.2	0.0%	estimated ± 5.1 pp, medium confidence
123	GPT-5.3 Codex	0.0%	estimated ± 5.1 pp, medium confidence
124	Claude Opus 4.5	0.0%	estimated ± 5.1 pp, medium confidence
125	Claude Sonnet 4.5	0.0%	estimated ± 5.1 pp, medium confidence
126	GPT-5.1-Codex	0.0%	estimated ± 5.1 pp, medium confidence
127	GPT-5.2-Codex	0.0%	estimated ± 5.1 pp, medium confidence
128	Claude 4.1 Opus	0.0%	estimated ± 5.1 pp, medium confidence
129	Qwen3.5 Plus	0.0%	estimated ± 5.1 pp, medium confidence
130	Claude 4 Sonnet	0.0%	estimated ± 5.1 pp, medium confidence
131	Gemini 3 Flash	0.0%	estimated ± 5.1 pp, low confidence
132	Gemini 3 Pro	0.0%	estimated ± 5.1 pp, low confidence
133	Kimi K2.5	0.0%	estimated ± 5.1 pp, low confidence
134	GPT-5 (high)	0.0%	estimated ± 5.1 pp, low confidence
135	Agents-A1	0.0%	estimated ± 6.1 pp, low confidence
136	Agents-A1-4B	0.0%	estimated ± 6.1 pp, low confidence
137	Claude Haiku 4.5	0.0%	measured
138	Claude Opus 4.8	0.0%	measured
139	Claude Sonnet 4.6	0.0%	measured
140	Claude Sonnet 5	0.0%	measured
141	DeepSeek V4 Flash 0731	0.0%	measured
142	dots3-note Preview	0.0%	estimated ± 6.1 pp, low confidence
143	Gemini 3.5 Flash	0.0%	measured
144	Gemini 3.6 Flash	0.0%	measured
145	GLM-5	0.0%	estimated ± 6.1 pp, low confidence
146	GLM-5.2	0.0%	measured
147	GLM-5.3	0.0%	measured
148	GLM-5.3-Flash	0.0%	measured
149	GPT-5.3-Codex-Spark	0.0%	measured
150	GPT-5.4	0.0%	measured
151	GPT-5.5	0.0%	measured
152	GPT-5.6 Luna	0.0%	measured
153	GPT-6 Luna	0.0%	measured
154	Grok 4.20	0.0%	measured
155	Ling 3.0 Flash	0.0%	estimated ± 4.6 pp, medium confidence
156	Ling 3.0 Flash Sante	0.0%	estimated ± 6.1 pp, low confidence
157	Llama 4 Maverick	0.0%	measured
158	LongCat-Flash-Lite-Sparse	0.0%	estimated ± 6.1 pp, low confidence
159	MiMo-V2.5-Pro	0.0%	measured
160	MiMo-V2.6-Flash	0.0%	estimated ± 3.0 pp, high confidence
161	MiniMax M3	0.0%	measured
162	Muse Spark 1.1	0.0%	estimated ± 3.0 pp, medium confidence
163	Muse Spark 1.2	0.0%	measured
164	Muse Spark 1.3	0.0%	measured
165	Muse Spark Open	0.0%	measured
166	Muse Wonder	0.0%	measured
167	Nemotron 3.5 Bolt 74B-A9B BF16	0.0%	estimated ± 6.1 pp, low confidence
168	Nemotron 3 Ultra	0.0%	measured
169	Ornith-1.5-35B-A3B	0.0%	estimated ± 6.1 pp, low confidence
170	Ornith-1.5-397B	0.0%	estimated ± 4.0 pp, high confidence
171	Ornith-1.5-9B	0.0%	estimated ± 6.1 pp, low confidence
172	Qwen3.5-27B	0.0%	estimated ± 6.1 pp, low confidence
173	Qwen3.5-35B-A3B	0.0%	estimated ± 6.1 pp, low confidence
174	Qwen3.5 397B	0.0%	estimated ± 6.1 pp, low confidence
175	Qwen3.8-27B	0.0%	measured
176	Qwen3.8 Max	0.0%	measured
177	Beam 1.0 501B-A23B FP8	0.0%	estimated ± 6.1 pp, low confidence
178	Beam 1.0 501B-A23B NVFP4	0.0%	estimated ± 6.1 pp, low confidence
179	Sage-A1.5	0.0%	estimated ± 6.1 pp, low confidence
180	Sarvam 105B	0.0%	estimated ± 6.1 pp, low confidence
181	Sarvam 30B	0.0%	estimated ± 6.1 pp, low confidence
182	Spark-X2.6-4B	0.0%	estimated ± 6.1 pp, low confidence

MazeBench leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General