Vision & documents

ScreenSpot Pro leaderboard

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

As of 2026-10-07, the highest measured score on ScreenSpot Pro is 92.7% by GPT-6 Astra. 108 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	92.7%	measured
2	Claude Mythos 5	89.8%	estimated ± 7.1 pp, medium confidence
3	Claude Opus 4.8	87.9%	measured
4	Claude Opus 5.5	87.8%	estimated ± 10.7 pp, low confidence
5	Claude Opus 4.7 (Adaptive)	87.3%	estimated ± 7.1 pp, medium confidence
6	GPT-6.1 Sol	85.9%	estimated ± 10.7 pp, low confidence
7	GLM-5.3-Flash	85.5%	estimated ± 7.1 pp, medium confidence
8	Gemini 3.8 Flash	85.4%	estimated ± 10.7 pp, low confidence
9	GPT-5.4	85.4%	measured
10	GPT-5.6 Sol	85.3%	estimated ± 10.0 pp, low confidence
11	LFM2.5-VL-3B	85.1%	estimated ± 2.5 pp, low confidence
12	Qwen3.6-27B	85.0%	estimated ± 2.5 pp, low confidence
13	Grok 4.20	85.0%	estimated ± 2.5 pp, low confidence
14	Qwen3.6-35B-A3B	85.0%	estimated ± 2.5 pp, low confidence
15	Gemini 3.7 Flash	84.7%	estimated ± 7.1 pp, medium confidence
16	dots3-note Preview	84.6%	estimated ± 2.5 pp, medium confidence
17	Qwen3.8 Max	84.5%	measured
18	Claude Opus 5	84.4%	estimated ± 10.7 pp, low confidence
19	Gemini 3.1 Pro	84.4%	measured
20	Muse Spark 1.1	84.3%	estimated ± 7.1 pp, medium confidence
21	Claude Sonnet 5	84.2%	estimated ± 7.1 pp, medium confidence
22	Muse Spark	84.1%	measured
23	Claude Opus 4.6	83.1%	measured
24	Gemini 3.6 Flash	82.7%	estimated ± 10.7 pp, low confidence
25	Kimi K3	82.4%	estimated ± 4.5 pp, medium confidence
26	GPT-6 Sol	82.4%	estimated ± 10.7 pp, low confidence
27	Qwen3.8 Max Preview	82.3%	estimated ± 10.7 pp, low confidence
28	Sakana Fugu-Ultra	82.0%	estimated ± 7.1 pp, medium confidence
29	Step 3.7 Flash	81.9%	estimated ± 2.5 pp, medium confidence
30	GPT-5.5	80.8%	estimated ± 10.0 pp, low confidence
31	Sakana Fugu	80.0%	estimated ± 7.1 pp, medium confidence
32	Grok 4.5	79.6%	estimated ± 10.7 pp, low confidence
33	GPT-5.6 Terra	79.5%	estimated ± 10.0 pp, low confidence
34	Seed 2.1 Pro	79.3%	estimated ± 4.5 pp, medium confidence
35	Qwen3.7 Plus	79.0%	measured
36	GPT-6 Luna	78.8%	estimated ± 10.7 pp, low confidence
37	Gemini 3.5 Flash	78.7%	estimated ± 7.1 pp, medium confidence
38	Apodex 1.1	78.2%	estimated ± 10.7 pp, low confidence
39	Apodex 1.1 Mini	78.2%	estimated ± 10.7 pp, low confidence
40	Gemini 3.5 Flash-Lite	78.0%	estimated ± 10.7 pp, low confidence
41	Ling 3.0 Flash VL	78.0%	estimated ± 10.7 pp, low confidence
42	Qwen3.8-Omni-Flash	77.8%	estimated ± 4.5 pp, medium confidence
43	Gemini 3 Flash	77.6%	estimated ± 10.7 pp, low confidence
44	GPT-5.3 Codex	77.5%	estimated ± 10.7 pp, low confidence
45	DeepSeek V4.1 Flash	75.8%	estimated ± 10.7 pp, low confidence
46	Qwen3.8-Flash-Next	75.6%	estimated ± 4.5 pp, medium confidence
47	Muse Glimmer 30B	75.4%	measured
48	Inkling	75.2%	estimated ± 7.1 pp, medium confidence
49	Claude Opus 4.7	75.2%	estimated ± 10.7 pp, low confidence
50	Mistral Large 4	75.2%	estimated ± 10.7 pp, low confidence
51	GPT-5.2-Codex	75.1%	estimated ± 10.7 pp, low confidence
52	Seed 2.1 Turbo	74.7%	estimated ± 4.5 pp, medium confidence
53	Qwen3.8-27B	74.5%	estimated ± 4.5 pp, medium confidence
54	GPT-5.1	74.2%	estimated ± 10.7 pp, low confidence
55	Kimi K2.5	74.2%	estimated ± 10.0 pp, low confidence
56	Kimi K2.5 (Reasoning)	74.2%	estimated ± 10.0 pp, low confidence
57	Claude Opus 4.6 (Adaptive)	74.1%	estimated ± 10.7 pp, low confidence
58	Inkling-Small	74.1%	estimated ± 7.1 pp, medium confidence
59	GPT-5.6 Luna	74.0%	estimated ± 10.0 pp, low confidence
60	Gemini 2.5 Pro	73.6%	estimated ± 10.7 pp, low confidence
61	MiMo-V2.5	73.6%	estimated ± 7.1 pp, medium confidence
62	Grok 4.3	73.3%	estimated ± 10.0 pp, low confidence
63	MiniMax M3	73.3%	estimated ± 10.0 pp, low confidence
64	Pareto 26.9	73.0%	estimated ± 10.0 pp, low confidence
65	GPT-5 (medium)	73.0%	estimated ± 10.7 pp, low confidence
66	GPT-5 (high)	72.9%	estimated ± 10.7 pp, low confidence
67	Gemini 3 Pro	72.7%	measured
68	Claude Opus 4.5 Thinking	72.7%	estimated ± 10.7 pp, low confidence
69	MiMo-V2.6-Flash	71.8%	estimated ± 10.7 pp, low confidence
70	GLM-5V-Turbo	71.5%	estimated ± 10.7 pp, low confidence
71	GPT-5.1-Codex	71.2%	estimated ± 10.7 pp, low confidence
72	GPT-5.1-Codex-Max	71.2%	estimated ± 10.7 pp, low confidence
73	Holo2-235B-A22B	70.6%	measured
74	Gemma 4 31B	70.4%	estimated ± 10.0 pp, low confidence
75	GPT-5.4 mini	69.7%	estimated ± 10.0 pp, low confidence
76	Kimi K2.6	69.7%	estimated ± 4.5 pp, medium confidence
77	o3	68.9%	estimated ± 10.7 pp, low confidence
78	MiMo-V2-Omni	68.7%	estimated ± 10.7 pp, low confidence
79	Step 5 Preview	68.3%	estimated ± 10.0 pp, low confidence
80	Qwen3.6 Plus	68.2%	measured
81	Grok 4	67.8%	estimated ± 10.7 pp, low confidence
82	Qwen3.5-122B-A10B	67.5%	estimated ± 4.5 pp, medium confidence
83	Qwen3.5-27B	67.1%	estimated ± 4.5 pp, medium confidence
84	Claude 4.1 Opus Thinking	67.0%	estimated ± 10.7 pp, low confidence
85	Claude Sonnet 4.6	67.0%	estimated ± 7.1 pp, medium confidence
86	Holo2-30B-A3B	66.1%	measured
87	Qwen3.5 397B	65.6%	measured
88	Gemini 2.5 Flash	65.1%	estimated ± 10.7 pp, low confidence
89	Mistral Medium 3.5 128B	64.7%	estimated ± 10.7 pp, low confidence
90	Grok 4.1 Fast (Reasoning)	63.6%	estimated ± 10.7 pp, low confidence
91	Gemma 4 26B A4B	63.3%	estimated ± 10.0 pp, low confidence
92	Qwen3.5-35B-A3B	63.3%	estimated ± 4.5 pp, medium confidence
93	Claude 4 Sonnet	63.0%	estimated ± 10.7 pp, low confidence
94	Llama 4 Maverick	62.8%	estimated ± 10.7 pp, low confidence
95	Grok 4 Fast (Reasoning)	62.6%	estimated ± 10.7 pp, low confidence
96	GPT-4.1	62.3%	estimated ± 10.7 pp, low confidence
97	Qwen3-Omni-30B-A3B-Thinking	61.7%	estimated ± 10.7 pp, low confidence
98	GPT-5.2	61.6%	estimated ± 4.5 pp, medium confidence
99	GPT-4.1 mini	60.9%	estimated ± 10.7 pp, low confidence
100	Mistral Small 4	60.1%	estimated ± 10.7 pp, low confidence
101	Mistral Small 4 (Reasoning)	60.1%	estimated ± 10.7 pp, low confidence
102	Mistral Large 3	59.6%	estimated ± 10.7 pp, low confidence
103	Qwen3-Omni-30B-A3B-Instruct	59.5%	estimated ± 10.7 pp, low confidence
104	Gemini 1.5 Pro	59.3%	estimated ± 10.7 pp, low confidence
105	Holo2-8B	58.9%	measured
106	Mistral Medium 3	58.6%	estimated ± 10.7 pp, low confidence
107	Llama 4 Scout	58.6%	estimated ± 10.7 pp, low confidence
108	Qwen3.5 397B (Reasoning)	58.5%	estimated ± 10.7 pp, low confidence
109	Gemini 3.1 Flash-Lite	58.3%	estimated ± 7.1 pp, medium confidence
110	Gemma 4 E4B	58.1%	estimated ± 10.7 pp, low confidence
111	Nemotron 3 Nano Omni 30B A3B	57.8%	measured
112	Grok 4.1 Fast	57.4%	estimated ± 10.7 pp, low confidence
113	Gemma 3 27B	57.3%	estimated ± 10.7 pp, low confidence
114	Interfaze Beta	57.3%	estimated ± 10.0 pp, low confidence
115	Holo2-4B	57.2%	measured
116	Gemma 4 E2B	56.7%	estimated ± 10.7 pp, low confidence
117	Nova Pro	56.7%	estimated ± 10.7 pp, low confidence
118	GPT-4o mini	56.4%	estimated ± 10.7 pp, low confidence
119	GPT-4.1 nano	56.2%	estimated ± 10.7 pp, low confidence
120	Claude 3 Haiku	55.8%	estimated ± 10.7 pp, low confidence
121	LFM2.5-VL-1.6B-Extract	55.7%	estimated ± 10.7 pp, low confidence
122	Phi-4 Multimodal Instruct	55.6%	estimated ± 10.7 pp, low confidence
123	Gemma 4 12B	55.5%	estimated ± 4.5 pp, medium confidence
124	GPT-5.4 nano	46.7%	estimated ± 10.0 pp, low confidence
125	Claude Opus 4.5	45.7%	measured
126	Command A+	15.3%	estimated ± 7.1 pp, low confidence

ScreenSpot Pro leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General