Agentic · tools

CyberGym leaderboard

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

As of 2026-10-07, the highest measured score on CyberGym is 95.1% by MiMo-V2.6-Flash. 151 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.8-Flash-Next	100.0%	estimated ± 5.4 pp, low confidence
2	MiMo-V2.6-Flash	95.1%	measured
3	Claude Opus 5.5	94.9%	estimated ± 2.4 pp, low confidence
4	MiMo-V2.6-Pro	94.0%	measured
5	Claude Fable 5	93.4%	estimated ± 5.0 pp, low confidence
6	Grok 4.7	92.1%	estimated ± 6.2 pp, low confidence
7	Claude Opus 5	89.1%	estimated ± 2.4 pp, low confidence
8	Gemini 4 Argon	88.8%	estimated ± 3.7 pp, high confidence
9	Claude Sonnet 5.5	88.7%	estimated ± 2.4 pp, low confidence
10	DeepSeek V4.1 Flash	88.1%	measured
11	Muse Spark 1.3	88.0%	estimated ± 3.7 pp, high confidence
12	Ling 3.1 Flash	87.9%	measured
13	Fugu Cyber	86.9%	measured
14	Atria Dawn Preview	86.5%	measured
15	Gemini 3.8 Flash Cyber	86.2%	measured
16	Gemini 3.8 Flash	85.6%	estimated ± 5.0 pp, medium confidence
17	Step 5 Preview	84.7%	measured
18	GLM-5.3	84.5%	measured
19	GPT-5.6 Sol	84.5%	measured
20	Holo3-35B-A3B	84.3%	estimated ± 11.3 pp, low confidence
21	Claude Mythos 5	83.8%	measured
22	Qwen3.8 Max Preview	83.6%	estimated ± 5.3 pp, medium confidence
23	DeepSeek V4 Pro 0813	83.3%	measured
24	Gemini 3.5 Flash Cyber	83.2%	measured
25	Claude Mythos Preview	83.1%	measured
26	GPT-5.5 Pro	82.8%	estimated ± 4.6 pp, high confidence
27	GPT-5.4 Pro	82.4%	estimated ± 4.6 pp, high confidence
28	GPT-6.1 Sol	82.1%	estimated ± 3.7 pp, high confidence
29	GPT-5.5	81.8%	measured
30	GPT-5.6 Terra	81.8%	measured
31	Qwen3.8-27B	81.3%	estimated ± 5.3 pp, medium confidence
32	GPT-6 Sol	80.8%	estimated ± 3.7 pp, high confidence
33	Claude Haiku 5.5	80.1%	estimated ± 2.4 pp, medium confidence
34	Claude Sonnet 5	80.1%	estimated ± 2.4 pp, medium confidence
35	Claude Fable 5.1	80.0%	estimated ± 3.7 pp, high confidence
36	GPT-6 Astra	79.9%	estimated ± 2.4 pp, medium confidence
37	Kimi K3	79.7%	estimated ± 3.7 pp, high confidence
38	Claude Opus 4.8	79.6%	estimated ± 4.6 pp, high confidence
39	Gemini 3.7 Flash	79.6%	estimated ± 3.7 pp, high confidence
40	GPT-6 Luna	79.5%	estimated ± 6.2 pp, medium confidence
41	Muse Spark 1.2	79.3%	estimated ± 5.3 pp, medium confidence
42	Qwen3.8 Max	79.3%	estimated ± 2.4 pp, medium confidence
43	Apodex 1.1	79.2%	estimated ± 2.4 pp, medium confidence
44	Ornith-1.5-397B	79.2%	estimated ± 2.4 pp, medium confidence
45	MiniMax M3	79.2%	estimated ± 4.6 pp, high confidence
46	Kimi K2.6	79.0%	estimated ± 4.6 pp, high confidence
47	GPT-5.4	79.0%	measured
48	GLM-5.3-Flash	78.8%	estimated ± 2.4 pp, medium confidence
49	Holo3-122B-A10B	78.8%	estimated ± 11.3 pp, low confidence
50	Mistral Large 4	78.8%	estimated ± 6.2 pp, medium confidence
51	Hy4 preview	78.4%	measured
52	Qwen3.7 Max	78.1%	estimated ± 2.4 pp, medium confidence
53	GPT-5.6 Luna	77.9%	measured
54	Grok 4.5	77.9%	estimated ± 5.3 pp, medium confidence
55	dots3-note Preview	77.8%	estimated ± 2.4 pp, medium confidence
56	Grok 4.6	77.1%	estimated ± 5.0 pp, medium confidence
57	Agents-A1	76.8%	estimated ± 2.4 pp, medium confidence
58	Step 3.7 Flash	76.8%	estimated ± 2.4 pp, medium confidence
59	DeepSeek V4 Flash 0731	76.7%	measured
60	Nemotron 3 Ultra	76.3%	estimated ± 2.4 pp, low confidence
61	Ornith-1.5-35B-A3B	76.3%	estimated ± 2.4 pp, low confidence
62	Ornith-1.5-9B	76.3%	estimated ± 2.4 pp, low confidence
63	GLM-5.2	75.8%	estimated ± 5.3 pp, medium confidence
64	UI-Mate-27B	75.6%	estimated ± 11.3 pp, low confidence
65	Inkling-Small	75.6%	estimated ± 4.6 pp, high confidence
66	Beam	75.6%	estimated ± 4.6 pp, high confidence
67	Inkling	75.4%	estimated ± 4.6 pp, high confidence
68	Claude Opus 4.7 (Adaptive)	73.1%	measured
69	Claude Opus 4.7	72.5%	estimated ± 5.0 pp, medium confidence
70	Ling 3.0 Flash	72.3%	estimated ± 4.6 pp, high confidence
71	Quasar 438B	70.7%	estimated ± 5.3 pp, medium confidence
72	Agents-A1-4B	68.8%	estimated ± 4.6 pp, medium confidence
73	Gemini 3.6 Flash	68.7%	estimated ± 5.3 pp, medium confidence
74	GLM-5.1	68.7%	measured
75	GPT-5.2	68.1%	estimated ± 4.6 pp, medium confidence
76	Qwen3.5-122B-A10B	66.7%	estimated ± 4.6 pp, medium confidence
77	Claude Opus 4.6	66.6%	measured
78	Gemini 3.5 Flash	66.5%	estimated ± 5.3 pp, medium confidence
79	Solar Open 2	66.1%	estimated ± 9.9 pp, medium confidence
80	Apodex 1.1 Mini	65.8%	estimated ± 6.2 pp, medium confidence
81	Gemini 3 Pro	65.7%	estimated ± 10.0 pp, low confidence
82	Qwen3.5 397B	65.5%	estimated ± 4.6 pp, medium confidence
83	Hy3	65.3%	estimated ± 5.3 pp, medium confidence
84	Hy3 Preview	65.3%	estimated ± 5.3 pp, medium confidence
85	Claude Sonnet 4.6	65.2%	measured
86	Qwen3.5-27B	64.8%	estimated ± 4.6 pp, medium confidence
87	Qwen3.5-35B-A3B	64.8%	estimated ± 4.6 pp, medium confidence
88	Kimi K2.5	64.5%	estimated ± 4.6 pp, medium confidence
89	Kimi K2.5 (Reasoning)	64.5%	estimated ± 4.6 pp, medium confidence
90	Ling 3.0 Flash VL	63.7%	estimated ± 6.2 pp, medium confidence
91	MiMo-V2.5-Pro	63.0%	estimated ± 5.3 pp, low confidence
92	Kimi K2.7 Code	62.8%	estimated ± 5.3 pp, low confidence
93	Ling 3.0 Flash FP8	61.7%	estimated ± 5.3 pp, low confidence
94	Qwen3.6-27B	61.0%	estimated ± 5.3 pp, low confidence
95	Qwen3.7 Plus	60.7%	estimated ± 5.3 pp, low confidence
96	GPT-5.4 mini	60.6%	estimated ± 5.3 pp, low confidence
97	GPT-5.4 nano	59.2%	estimated ± 5.3 pp, low confidence
98	Muse Spark 1.1	59.0%	measured
99	Grok 4.3	58.8%	estimated ± 5.3 pp, low confidence
100	MiniMax M2.7	58.5%	estimated ± 5.3 pp, low confidence
101	LLaDA2.2-flash	58.4%	estimated ± 9.9 pp, medium confidence
102	GLM-4.7	58.0%	estimated ± 4.6 pp, medium confidence
103	Gemini 3.5 Flash-Lite	57.8%	estimated ± 5.3 pp, low confidence
104	Qwen3.6-35B-A3B	57.1%	estimated ± 5.3 pp, low confidence
105	GPT-5.3 Codex	56.4%	estimated ± 10.0 pp, low confidence
106	Solar Pro 4	55.7%	estimated ± 4.6 pp, medium confidence
107	LongCat-Flash-Lite-Sparse	55.2%	estimated ± 4.6 pp, medium confidence
108	Gemini 3 Flash	55.1%	estimated ± 10.0 pp, low confidence
109	Muse Glimmer 30B	53.6%	estimated ± 5.3 pp, low confidence
110	Gemini 3.1 Pro	53.5%	estimated ± 5.3 pp, low confidence
111	Mistral Medium 3.5 128B	52.7%	estimated ± 5.3 pp, low confidence
112	UI-Mate-9B	52.4%	estimated ± 11.3 pp, low confidence
113	Qwen3.6 Plus	51.7%	estimated ± 6.2 pp, low confidence
114	Gemma 4 31B	50.7%	estimated ± 5.3 pp, low confidence
115	Claude Opus 4.5	50.6%	measured
116	GPT-OSS 120B	50.3%	estimated ± 5.3 pp, low confidence
117	A.X K2	49.3%	estimated ± 6.2 pp, low confidence
118	Nemotron 3 Super 100B	48.7%	estimated ± 5.3 pp, low confidence
119	Granite 4.2 8B	48.4%	estimated ± 5.3 pp, low confidence
120	Command A+	48.3%	estimated ± 5.3 pp, low confidence
121	Gemini 2.5 Pro	48.3%	estimated ± 5.3 pp, low confidence
122	GPT-5.2-Codex	47.9%	estimated ± 10.0 pp, low confidence
123	Claude 4.1 Opus	47.6%	estimated ± 11.4 pp, low confidence
124	Mistral Large 3	47.4%	estimated ± 5.3 pp, low confidence
125	Mistral Small 4	46.7%	estimated ± 5.3 pp, low confidence
126	Mistral Small 4 (Reasoning)	46.7%	estimated ± 5.3 pp, low confidence
127	GPT-OSS 20B	46.6%	estimated ± 5.3 pp, low confidence
128	Trinity-Large-Preview	46.4%	estimated ± 5.3 pp, low confidence
129	Trinity-Large-Thinking	46.4%	estimated ± 5.3 pp, low confidence
130	Nemotron 3 Nano 30B	46.3%	estimated ± 5.3 pp, low confidence
131	DeepSeek V3	46.2%	estimated ± 5.3 pp, low confidence
132	Celeris-1	46.1%	estimated ± 5.3 pp, low confidence
133	Llama 4 Maverick	46.0%	estimated ± 5.3 pp, low confidence
134	Llama 4 Scout	46.0%	estimated ± 5.3 pp, low confidence
135	GPT-5 (high)	45.7%	estimated ± 6.2 pp, low confidence
136	Gemma 3 27B	45.7%	estimated ± 5.3 pp, low confidence
137	GPT-5.1-Codex	44.9%	estimated ± 10.0 pp, low confidence
138	Nemotron 3.5 Lightning 30B A3B NVFP4	44.8%	estimated ± 4.6 pp, medium confidence
139	Grok Build 0.1	44.2%	estimated ± 10.0 pp, low confidence
140	Muse Spark	43.5%	measured
141	Claude Sonnet 4.5	43.3%	estimated ± 10.0 pp, low confidence
142	GLM-5	43.2%	measured
143	Qwen3.5 Plus	42.2%	estimated ± 11.4 pp, low confidence
144	Grok 4.1 Fast	41.6%	estimated ± 10.0 pp, low confidence
145	MiMo-V2.5	41.1%	estimated ± 10.0 pp, low confidence
146	Claude Haiku 4.5	37.9%	estimated ± 11.4 pp, low confidence
147	GPT-5.1	37.8%	estimated ± 6.2 pp, low confidence
148	Qwen3 Max	36.9%	estimated ± 10.0 pp, low confidence
149	Grok 4	35.1%	estimated ± 10.0 pp, low confidence
150	Claude 4 Sonnet	31.7%	estimated ± 10.0 pp, low confidence
151	Gemini 3.1 Flash-Lite	30.2%	estimated ± 10.0 pp, low confidence
152	Grok 4.20	30.1%	estimated ± 10.0 pp, low confidence
153	MiMo-V2-Pro	28.0%	estimated ± 10.0 pp, low confidence
154	MiniCPM5-2B	25.8%	estimated ± 6.2 pp, low confidence
155	Laguna S 2.1	22.5%	estimated ± 6.0 pp, low confidence
156	GLM-5V-Turbo	21.1%	estimated ± 10.0 pp, low confidence
157	DeepSeek V3.2	19.8%	estimated ± 10.0 pp, low confidence
158	MiMo-V2-Flash	17.1%	estimated ± 6.2 pp, low confidence
159	GPT-4.1	15.6%	estimated ± 10.0 pp, low confidence
160	Granite 4.2 30B	10.4%	estimated ± 6.2 pp, low confidence
161	Gemma 4 26B A4B	9.2%	estimated ± 6.2 pp, low confidence
162	Ling 3.0 Tiny	8.7%	estimated ± 6.2 pp, low confidence
163	DeepSeek V3 0324	0.0%	estimated ± 6.2 pp, low confidence
164	Gemma 4 12B	0.0%	estimated ± 6.2 pp, low confidence
165	Gemma 4 E2B	0.0%	estimated ± 6.2 pp, low confidence
166	Gemma 4 E4B	0.0%	estimated ± 6.2 pp, low confidence
167	GPT-4.1 mini	0.0%	estimated ± 6.2 pp, low confidence
168	GPT-4.1 nano	0.0%	estimated ± 6.2 pp, low confidence
169	GPT-4o	0.0%	estimated ± 6.2 pp, low confidence
170	GPT-4o mini	0.0%	estimated ± 6.2 pp, low confidence
171	Granite 4.2 3B	0.0%	estimated ± 6.2 pp, low confidence
172	K-Exaone	0.0%	estimated ± 6.2 pp, low confidence
173	LFM2.5-2.6B	0.0%	estimated ± 6.2 pp, low confidence
174	Ling 2.6 Flash	0.0%	estimated ± 6.2 pp, low confidence
175	Mercury 2.5	0.0%	estimated ± 6.2 pp, low confidence
176	Nemotron 3 Nano Omni 30B A3B	0.0%	estimated ± 6.2 pp, low confidence
177	North Mini Code	0.0%	estimated ± 6.2 pp, low confidence
178	Solar Pro 3	0.0%	estimated ± 6.2 pp, low confidence
179	Ultravox v0.6 Llama 3.3 70B	0.0%	estimated ± 6.2 pp, low confidence

CyberGym leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General