Vision & documents

CharXiv leaderboard

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

As of 2026-10-07, the highest measured score on CharXiv is 93.5% by Claude Mythos 5. 91 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Mythos 5	93.5%	measured
2	Qwen3.8 Max	93.5%	measured
3	Qwen3.8-Omni-Flash	91.4%	measured
4	Gemini 3.8 Flash	91.4%	estimated ± 2.5 pp, high confidence
5	Kimi K3	91.3%	measured
6	Claude Opus 4.7 (Adaptive)	91.0%	measured
7	Qwen3.8-Flash-Next	90.6%	measured
8	Qwen3.8-27B	90.2%	measured
9	Claude Opus 4.8	89.9%	measured
10	dots3-note Preview	89.5%	estimated ± 2.5 pp, high confidence
11	GLM-5.3-Flash	89.4%	measured
12	Gemini 3.7 Flash	88.7%	measured
13	GPT-6 Astra	88.5%	estimated ± 3.9 pp, medium confidence
14	Muse Spark 1.1	88.4%	measured
15	Claude Sonnet 5	88.3%	measured
16	Sakana Fugu-Ultra	86.6%	measured
17	Muse Spark	86.4%	measured
18	GPT-5.6 Sol	86.2%	estimated ± 5.4 pp, medium confidence
19	Claude Opus 5.5	86.0%	estimated ± 7.5 pp, low confidence
20	GPT-6.1 Sol	86.0%	estimated ± 7.5 pp, low confidence
21	Claude Opus 5	86.0%	estimated ± 7.5 pp, low confidence
22	Gemini 3.6 Flash	86.0%	estimated ± 7.5 pp, low confidence
23	GPT-6 Sol	86.0%	estimated ± 7.5 pp, low confidence
24	Qwen3.8 Max Preview	86.0%	estimated ± 7.5 pp, low confidence
25	Qwen3.7 Plus	85.9%	measured
26	Grok 4.5	85.8%	estimated ± 7.5 pp, low confidence
27	GPT-6 Luna	85.7%	estimated ± 7.5 pp, low confidence
28	Claude Opus 4.6	85.7%	estimated ± 3.9 pp, high confidence
29	Apodex 1.1	85.6%	estimated ± 7.5 pp, low confidence
30	Apodex 1.1 Mini	85.6%	estimated ± 7.5 pp, low confidence
31	Gemini 3.5 Flash-Lite	85.6%	estimated ± 7.5 pp, low confidence
32	Ling 3.0 Flash VL	85.6%	estimated ± 7.5 pp, low confidence
33	Gemini 3 Flash	85.4%	estimated ± 7.5 pp, low confidence
34	Seed 2.1 Pro	85.4%	measured
35	GPT-5.3 Codex	85.4%	estimated ± 7.5 pp, low confidence
36	Sakana Fugu	85.1%	measured
37	GPT-5.5	84.5%	estimated ± 5.4 pp, medium confidence
38	DeepSeek V4.1 Flash	84.4%	estimated ± 7.5 pp, low confidence
39	Gemini 3.5 Flash	84.2%	measured
40	GPT-5.6 Terra	84.0%	estimated ± 5.4 pp, medium confidence
41	Claude Opus 4.7	83.8%	estimated ± 7.5 pp, low confidence
42	Mistral Large 4	83.8%	estimated ± 7.5 pp, low confidence
43	GPT-5.2-Codex	83.7%	estimated ± 7.5 pp, low confidence
44	GPT-5.4	82.8%	measured
45	Seed 2.1 Turbo	82.5%	measured
46	GPT-5.1	82.4%	estimated ± 7.5 pp, low confidence
47	Claude Opus 4.6 (Adaptive)	82.3%	estimated ± 7.5 pp, low confidence
48	GPT-5.2	82.1%	measured
49	Inkling	82.0%	measured
50	Kimi K2.5	81.6%	estimated ± 5.4 pp, medium confidence
51	Kimi K2.5 (Reasoning)	81.6%	estimated ± 5.4 pp, medium confidence
52	Qwen3.6 Plus	81.5%	measured
53	GPT-5.6 Luna	81.5%	estimated ± 5.4 pp, medium confidence
54	Gemini 3 Pro	81.4%	measured
55	Holo2-235B-A22B	81.4%	estimated ± 3.9 pp, high confidence
56	Inkling-Small	81.3%	measured
57	Gemini 2.5 Pro	81.3%	estimated ± 7.5 pp, low confidence
58	Qwen3.5-27B	81.2%	estimated ± 3.3 pp, high confidence
59	Grok 4.3	81.1%	estimated ± 5.4 pp, medium confidence
60	MiMo-V2.5	81.0%	measured
61	Pareto 26.9	81.0%	estimated ± 5.4 pp, medium confidence
62	Qwen3.5 397B	80.8%	measured
63	Kimi K2.6	80.4%	measured
64	Step 3.7 Flash	80.3%	estimated ± 4.9 pp, high confidence
65	Gemini 3.1 Pro	80.2%	measured
66	GPT-5 (medium)	79.9%	estimated ± 7.5 pp, low confidence
67	GPT-5 (high)	79.7%	estimated ± 7.5 pp, low confidence
68	Gemma 4 31B	79.6%	estimated ± 5.4 pp, medium confidence
69	Holo2-30B-A3B	79.6%	estimated ± 3.9 pp, high confidence
70	GPT-5.4 mini	79.2%	estimated ± 5.4 pp, medium confidence
71	Claude Opus 4.5 Thinking	79.2%	estimated ± 7.5 pp, low confidence
72	Muse Glimmer 30B	78.8%	measured
73	Qwen3.5-35B-A3B	78.7%	estimated ± 3.3 pp, high confidence
74	Step 5 Preview	78.4%	estimated ± 5.4 pp, medium confidence
75	Qwen3.6-27B	78.4%	measured
76	Qwen3.6-35B-A3B	78.0%	measured
77	Claude Sonnet 4.6	77.4%	measured
78	Qwen3.5-122B-A10B	77.2%	measured
79	MiMo-V2.6-Flash	76.9%	estimated ± 7.5 pp, low confidence
80	Holo2-8B	76.3%	estimated ± 3.9 pp, high confidence
81	Nemotron 3 Nano Omni 30B A3B	76.3%	measured
82	GLM-5V-Turbo	76.2%	estimated ± 7.5 pp, low confidence
83	GPT-5.1-Codex	75.5%	estimated ± 7.5 pp, low confidence
84	GPT-5.1-Codex-Max	75.5%	estimated ± 7.5 pp, low confidence
85	Holo2-4B	75.5%	estimated ± 3.9 pp, high confidence
86	Gemma 4 26B A4B	75.3%	estimated ± 5.4 pp, medium confidence
87	Gemma 4 12B	73.9%	estimated ± 3.3 pp, high confidence
88	Gemini 3.1 Flash-Lite	73.2%	measured
89	o3	71.9%	estimated ± 7.5 pp, low confidence
90	MiniMax M3	71.8%	estimated ± 2.7 pp, low confidence
91	MiMo-V2-Omni	71.7%	estimated ± 7.5 pp, low confidence
92	Grok 4	71.0%	estimated ± 7.5 pp, low confidence
93	Interfaze Beta	70.8%	estimated ± 5.4 pp, medium confidence
94	Claude 4.1 Opus Thinking	70.7%	estimated ± 7.5 pp, low confidence
95	Gemini 2.5 Flash	70.4%	estimated ± 7.5 pp, low confidence
96	Mistral Medium 3.5 128B	70.4%	estimated ± 7.5 pp, low confidence
97	Grok 4.1 Fast (Reasoning)	70.4%	estimated ± 7.5 pp, low confidence
98	Claude 4 Sonnet	70.4%	estimated ± 7.5 pp, low confidence
99	Llama 4 Maverick	70.4%	estimated ± 7.5 pp, low confidence
100	Grok 4 Fast (Reasoning)	70.4%	estimated ± 7.5 pp, low confidence
101	GPT-4.1	70.4%	estimated ± 7.5 pp, low confidence
102	Qwen3-Omni-30B-A3B-Thinking	70.3%	estimated ± 7.5 pp, low confidence
103	GPT-4.1 mini	70.3%	estimated ± 7.5 pp, low confidence
104	Mistral Small 4	70.3%	estimated ± 7.5 pp, low confidence
105	Mistral Small 4 (Reasoning)	70.3%	estimated ± 7.5 pp, low confidence
106	Gemini 1.5 Pro	70.3%	estimated ± 7.5 pp, low confidence
107	Mistral Large 3	70.3%	estimated ± 7.5 pp, low confidence
108	Qwen3-Omni-30B-A3B-Instruct	70.3%	estimated ± 7.5 pp, low confidence
109	Claude 3 Haiku	70.3%	estimated ± 7.5 pp, low confidence
110	Gemma 3 27B	70.3%	estimated ± 7.5 pp, low confidence
111	Gemma 4 E2B	70.3%	estimated ± 7.5 pp, low confidence
112	Gemma 4 E4B	70.3%	estimated ± 7.5 pp, low confidence
113	GPT-4.1 nano	70.3%	estimated ± 7.5 pp, low confidence
114	GPT-4o mini	70.3%	estimated ± 7.5 pp, low confidence
115	Grok 4.1 Fast	70.3%	estimated ± 7.5 pp, low confidence
116	LFM2.5-VL-1.6B-Extract	70.3%	estimated ± 7.5 pp, low confidence
117	Llama 4 Scout	70.3%	estimated ± 7.5 pp, low confidence
118	Mistral Medium 3	70.3%	estimated ± 7.5 pp, low confidence
119	Nova Pro	70.3%	estimated ± 7.5 pp, low confidence
120	Phi-4 Multimodal Instruct	70.3%	estimated ± 7.5 pp, low confidence
121	Qwen3.5 397B (Reasoning)	70.3%	estimated ± 7.5 pp, low confidence
122	Claude Opus 4.5	68.5%	measured
123	Ternary Bonsai 2 27B	65.0%	estimated ± 4.1 pp, low confidence
124	GPT-5.4 nano	61.0%	estimated ± 5.4 pp, medium confidence
125	Grok 4.20	60.9%	measured
126	Command A+	52.7%	measured
127	LFM2.5-VL-3B	41.4%	estimated ± 4.1 pp, low confidence
128	ZAYA1-VL-8B	14.1%	estimated ± 4.1 pp, low confidence
129	North Micro Vision Instruct	4.7%	estimated ± 4.1 pp, low confidence
130	LFM2.5-VL-450M	0.0%	estimated ± 4.1 pp, low confidence

CharXiv leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General