Vision & documents

RealWorldQA leaderboard

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

As of 2026-10-07, the highest measured score on RealWorldQA is 88.5% by Qwen3.8-Flash-Next. 114 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	91.4%	estimated ± 1.2 pp, low confidence
2	GPT-6 Astra	91.0%	estimated ± 1.2 pp, low confidence
3	GPT-6.1 Sol	90.6%	estimated ± 1.2 pp, low confidence
4	Gemini 3.8 Flash	90.4%	estimated ± 1.2 pp, low confidence
5	Claude Opus 5	90.0%	estimated ± 1.2 pp, low confidence
6	GPT-5.6 Sol	89.3%	estimated ± 1.2 pp, low confidence
7	Gemini 3.6 Flash	89.2%	estimated ± 1.2 pp, low confidence
8	GPT-6 Sol	89.1%	estimated ± 1.2 pp, low confidence
9	Qwen3.8 Max Preview	89.0%	estimated ± 1.2 pp, low confidence
10	Qwen3.8-Flash-Next	88.5%	measured
11	Kimi K3	88.2%	estimated ± 0.8 pp, medium confidence
12	Seed 2.1 Pro	88.2%	estimated ± 0.8 pp, medium confidence
13	Claude Mythos 5	88.0%	estimated ± 1.1 pp, medium confidence
14	Qwen3.8 Max	88.0%	measured
15	GPT-5.6 Terra	88.0%	estimated ± 1.2 pp, low confidence
16	Grok 4.5	87.8%	estimated ± 1.2 pp, medium confidence
17	Qwen3.8-Omni-Flash	87.7%	measured
18	GPT-5.5	87.6%	estimated ± 1.2 pp, medium confidence
19	Claude Opus 4.7 (Adaptive)	87.5%	estimated ± 1.1 pp, medium confidence
20	GPT-6 Luna	87.5%	estimated ± 1.2 pp, medium confidence
21	Claude Opus 4.8	87.3%	estimated ± 1.1 pp, medium confidence
22	Apodex 1.1	87.2%	estimated ± 1.2 pp, medium confidence
23	Apodex 1.1 Mini	87.2%	estimated ± 1.2 pp, medium confidence
24	GLM-5.3-Flash	87.2%	estimated ± 1.1 pp, medium confidence
25	Gemini 3.5 Flash-Lite	87.1%	estimated ± 1.2 pp, medium confidence
26	Ling 3.0 Flash VL	87.1%	estimated ± 1.2 pp, medium confidence
27	Gemini 3.1 Pro	87.0%	estimated ± 1.1 pp, medium confidence
28	Gemini 3.7 Flash	87.0%	estimated ± 1.1 pp, medium confidence
29	Muse Spark 1.1	87.0%	estimated ± 1.1 pp, medium confidence
30	Claude Sonnet 5	86.9%	estimated ± 1.1 pp, medium confidence
31	Qwen3.7 Plus	86.9%	measured
32	Gemini 3 Flash	86.9%	estimated ± 1.2 pp, medium confidence
33	GPT-5.6 Luna	86.9%	estimated ± 1.2 pp, medium confidence
34	MiniMax M3	86.9%	estimated ± 1.2 pp, medium confidence
35	GPT-5.3 Codex	86.8%	estimated ± 1.2 pp, medium confidence
36	Grok 4.3	86.6%	estimated ± 1.2 pp, medium confidence
37	Sakana Fugu-Ultra	86.6%	estimated ± 1.1 pp, medium confidence
38	Seed 2.1 Turbo	86.5%	estimated ± 0.8 pp, medium confidence
39	Sakana Fugu	86.3%	estimated ± 1.1 pp, medium confidence
40	Pareto 26.9	86.1%	estimated ± 4.8 pp, medium confidence
41	Gemini 3.5 Flash	86.1%	estimated ± 1.1 pp, medium confidence
42	DeepSeek V4.1 Flash	86.0%	estimated ± 1.2 pp, medium confidence
43	Qwen3.8-27B	85.9%	measured
44	Claude Opus 4.7	85.7%	estimated ± 1.2 pp, medium confidence
45	Mistral Large 4	85.7%	estimated ± 1.2 pp, medium confidence
46	Step 5 Preview	85.7%	estimated ± 1.2 pp, medium confidence
47	GPT-5.4	85.7%	estimated ± 1.1 pp, medium confidence
48	GPT-5.2-Codex	85.7%	estimated ± 1.2 pp, medium confidence
49	Inkling	85.6%	estimated ± 1.1 pp, medium confidence
50	Inkling-Small	85.4%	estimated ± 1.1 pp, medium confidence
51	Muse Spark	85.4%	estimated ± 1.1 pp, medium confidence
52	MiMo-V2.5	85.4%	estimated ± 1.1 pp, medium confidence
53	Qwen3.6-35B-A3B	85.3%	measured
54	GPT-5.1	85.2%	estimated ± 1.2 pp, medium confidence
55	Claude Opus 4.6 (Adaptive)	85.2%	estimated ± 1.2 pp, medium confidence
56	Kimi K2.5	85.2%	estimated ± 1.2 pp, medium confidence
57	Kimi K2.5 (Reasoning)	85.2%	estimated ± 1.2 pp, medium confidence
58	Step 3.7 Flash	85.1%	estimated ± 1.2 pp, medium confidence
59	Muse Glimmer 30B	84.9%	estimated ± 1.1 pp, medium confidence
60	Gemini 2.5 Pro	84.9%	estimated ± 1.2 pp, medium confidence
61	Command A+	84.7%	estimated ± 0.8 pp, medium confidence
62	Nemotron 3 Nano Omni 30B A3B	84.7%	estimated ± 0.8 pp, medium confidence
63	Claude Sonnet 4.6	84.6%	estimated ± 1.1 pp, low confidence
64	GPT-5 (medium)	84.6%	estimated ± 1.2 pp, low confidence
65	GPT-5 (high)	84.5%	estimated ± 1.2 pp, low confidence
66	Claude Opus 4.5 Thinking	84.4%	estimated ± 1.2 pp, low confidence
67	Interfaze Beta	84.2%	estimated ± 4.8 pp, medium confidence
68	Qwen3.6-27B	84.1%	measured
69	Gemma 4 31B	84.1%	estimated ± 1.2 pp, low confidence
70	GPT-5.4 mini	84.0%	estimated ± 1.2 pp, low confidence
71	MiMo-V2.6-Flash	83.9%	estimated ± 1.2 pp, low confidence
72	GLM-5V-Turbo	83.7%	estimated ± 1.2 pp, low confidence
73	Gemini 3.1 Flash-Lite	83.7%	estimated ± 1.1 pp, low confidence
74	GPT-5.1-Codex	83.5%	estimated ± 1.2 pp, low confidence
75	GPT-5.1-Codex-Max	83.5%	estimated ± 1.2 pp, low confidence
76	o3	82.1%	estimated ± 1.2 pp, low confidence
77	MiMo-V2-Omni	82.0%	estimated ± 1.2 pp, low confidence
78	Gemma 4 26B A4B	81.6%	estimated ± 1.2 pp, low confidence
79	Grok 4	81.4%	estimated ± 1.2 pp, low confidence
80	Claude 4.1 Opus Thinking	80.8%	estimated ± 1.2 pp, low confidence
81	Ternary Bonsai 2 27B	80.1%	measured
82	Gemini 2.5 Flash	79.3%	estimated ± 1.2 pp, low confidence
83	GPT-5.4 nano	79.3%	estimated ± 1.2 pp, low confidence
84	Mistral Medium 3.5 128B	79.0%	estimated ± 1.2 pp, low confidence
85	Grok 4.1 Fast (Reasoning)	77.9%	estimated ± 1.2 pp, low confidence
86	Claude 4 Sonnet	77.3%	estimated ± 1.2 pp, low confidence
87	Grok 4.20	77.3%	estimated ± 1.1 pp, low confidence
88	Llama 4 Maverick	77.1%	estimated ± 1.2 pp, low confidence
89	Grok 4 Fast (Reasoning)	76.9%	estimated ± 1.2 pp, low confidence
90	GPT-4.1	76.5%	estimated ± 1.2 pp, low confidence
91	Qwen3-Omni-30B-A3B-Thinking	75.9%	estimated ± 1.2 pp, low confidence
92	GPT-4.1 mini	74.8%	estimated ± 1.2 pp, low confidence
93	Claude Opus 4.6	73.8%	estimated ± 1.1 pp, low confidence
94	Mistral Small 4	73.5%	estimated ± 1.2 pp, low confidence
95	Mistral Small 4 (Reasoning)	73.5%	estimated ± 1.2 pp, low confidence
96	LFM2.5-VL-3B	73.1%	measured
97	Mistral Large 3	72.7%	estimated ± 1.2 pp, low confidence
98	Qwen3-Omni-30B-A3B-Instruct	72.5%	estimated ± 1.2 pp, low confidence
99	Gemini 1.5 Pro	72.2%	estimated ± 1.2 pp, low confidence
100	Mistral Medium 3	70.7%	estimated ± 1.2 pp, low confidence
101	Llama 4 Scout	70.6%	estimated ± 1.2 pp, low confidence
102	Qwen3.5 397B (Reasoning)	70.4%	estimated ± 1.2 pp, low confidence
103	Gemma 4 E4B	69.4%	estimated ± 1.2 pp, low confidence
104	Grok 4.1 Fast	67.0%	estimated ± 1.2 pp, low confidence
105	Gemma 3 27B	66.7%	estimated ± 1.2 pp, low confidence
106	ZAYA1-VL-8B	65.0%	measured
107	Gemma 4 E2B	63.8%	estimated ± 1.2 pp, low confidence
108	Nova Pro	63.5%	estimated ± 1.2 pp, low confidence
109	Qwen3.5 397B	63.2%	estimated ± 0.8 pp, low confidence
110	North Micro Vision Instruct	62.2%	measured
111	GPT-4o mini	61.0%	estimated ± 1.2 pp, low confidence
112	GPT-4.1 nano	59.7%	estimated ± 1.2 pp, low confidence
113	LFM2.5-VL-450M	58.4%	measured
114	Claude 3 Haiku	50.0%	estimated ± 1.2 pp, low confidence
115	LFM2.5-VL-1.6B-Extract	44.9%	estimated ± 1.2 pp, low confidence
116	Qwen3.6 Plus	38.1%	estimated ± 0.8 pp, low confidence
117	Phi-4 Multimodal Instruct	28.0%	estimated ± 1.2 pp, low confidence
118	dots3-note Preview	26.0%	estimated ± 0.8 pp, low confidence
119	Kimi K2.6	16.4%	estimated ± 0.8 pp, low confidence
120	Gemini 3 Pro	3.9%	estimated ± 0.8 pp, low confidence
121	Qwen3.5-122B-A10B	1.8%	estimated ± 0.8 pp, low confidence
122	Qwen3.5-27B	1.2%	estimated ± 0.8 pp, low confidence
123	Qwen3.5-35B-A3B	0.0%	estimated ± 0.8 pp, low confidence
124	GPT-5.2	0.0%	estimated ± 0.8 pp, low confidence
125	Claude Opus 4.5	0.0%	estimated ± 0.8 pp, low confidence
126	Gemma 4 12B	0.0%	estimated ± 0.8 pp, low confidence

RealWorldQA leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General