Skip to content

Benchmark

GAIA

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
71
Scores recorded
291
Sources
3
IQ weight
1.01

Score distribution

37%
Mean
32.2%
Median
28.6%
Std dev
2.3%
10th percentile
80.6%
90th percentile
0%
Min
0%94.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.01 for chat models, measured over 71 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51DeepSeek V4 Pro· GAIA Leaderboard data71.8%
52Claude 3.5 SonnetLevel 1· GAIA Leaderboard data71.7%
53Claude Sonnet 3.7Level 1· GAIA Leaderboard data71.7%
54Gemini 3 Pro PreviewLevel 3· GAIA Leaderboard data71.4%
55GPT-6.1 SolLevel 3· GAIA Leaderboard data71.4%
56GPT 5.6Level 2· GAIA Leaderboard data71.1%
57DeepSeek V4 ProLevel 2· GAIA Leaderboard data70.4%
58GPT-5.6 TerraLevel 2· GAIA Leaderboard data69.2%
59GPT-5.6 Terra· GAIA Leaderboard data68.8%
60GPT 5.4· GAIA Leaderboard data68.1%
61o1Level 1· GAIA Leaderboard data67.9%
62Claude Sonnet 4Level 1· GAIA Leaderboard data67.7%
63DeepSeek V4 Flash 0731· GAIA Leaderboard data66.4%
64DeepSeek V4 Flash 0731Level 2· GAIA Leaderboard data66%
65GLM 5.2Level 3· GAIA Leaderboard data65.3%
66GLM 5.3 FlashLevel 1· GAIA Leaderboard data64.5%
67GPT-5.4 mini· GAIA Leaderboard data64.5%
68Gemini 3.1 Pro PreviewLevel 1, openrouter· GAIA Leaderboard data63.4%
69GLM 5.1Level 1· GAIA Leaderboard data63.4%
70Kimi K3Level 3· GAIA Leaderboard data63.3%
71GPT 4.1Level 1· GAIA Leaderboard data62.4%
72GPT-5.4 miniLevel 2· GAIA Leaderboard data61.6%
73o4 MiniLevel 1· GAIA Leaderboard data61.3%
74GPT 5.4Level 2· GAIA Leaderboard data61%
75GPT-5.6 Luna· GAIA Leaderboard data60.8%
76claudeLevel 2· GAIA Leaderboard data60.4%
77GPT-5.6 LunaLevel 2· GAIA Leaderboard data60.4%
78GPT 5.6Level 3· GAIA Leaderboard data59.2%
79GPT 4oLevel 1· GAIA Leaderboard data58.5%
80Gemini 2.5 ProLevel 3· GAIA Leaderboard data57.7%
81DeepSeek Flash· GAIA Leaderboard data57.5%
82Gemini 3 FlashLevel 1· GAIA Leaderboard data57%
83DeepSeek V3.2Level 1· GAIA Leaderboard data55.9%
84Claude Sonnet 3.7· GAIA Leaderboard data55.8%
85Claude 3.5 Sonnet· GAIA Leaderboard data55.2%
86o1· GAIA Leaderboard data55.2%
87DeepSeek V4 Flash· GAIA Leaderboard data55.1%
88Claude Opus 4.5Level 3· GAIA Leaderboard data55.1%
89GPT 5.4Level 3· GAIA Leaderboard data55.1%
90Claude Sonnet 4.6· GAIA Leaderboard data54.5%
91Claude 3.5 SonnetLevel 2· GAIA Leaderboard data53.5%
92Claude Sonnet 3.7Level 2· GAIA Leaderboard data53.5%
93o1Level 2· GAIA Leaderboard data53.5%
94DeepSeek V4 ProLevel 3· GAIA Leaderboard data53.1%
95DeepSeek V4.1 FlashLevel 1· GAIA Leaderboard data52.7%
96GPT 4oLevel 1, gpt-4o-2024-08-06· GAIA Leaderboard data51.6%
97Claude Sonnet 4· GAIA Leaderboard data51.2%
98DeepSeek V4 Flash 0731Level 3· GAIA Leaderboard data51%
99Qwen3.8 FlashLevel 3· GAIA Leaderboard data51%
100DeepSeek FlashLevel 2· GAIA Leaderboard data50.9%

Showing 51–100 of 291 results.

Data credits: Steel.dev Agent Leaderboards, MIT (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.