Skip to content

Benchmark

Gaia2

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

Models measured
13
Scores recorded
18
Sources
3
IQ weight
1.07

Score distribution

23.4%
Mean
21.9%
Median
12.6%
Std dev
7.4%
10th percentile
42.1%
90th percentile
4.4%
Min
4.4%43.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.07 for chat models, measured over 13 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Muse Glimmer 30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim43.3%
2Muse Glimmer 30Bhigh reasoning· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.43.3%
3GPT 5high· Gaia2 (Meta ARE) data42.1%
4Claude Sonnet 4thinking· Gaia2 (Meta ARE) data37.8%
5Claude Sonnet 4· Gaia2 (Meta ARE) data34.8%
6GPT 5low· Gaia2 (Meta ARE) data34.6%
7Gemini 2.5 Pro· Gaia2 (Meta ARE) data25.8%
8DeepSeek V3.1 Terminus· Gaia2 (Meta ARE) data23.1%
9DeepSeek V3.1· Gaia2 (Meta ARE) data21.9%
10Kimi K2· Gaia2 (Meta ARE) data20.1%
11GPT 5minimal· Gaia2 (Meta ARE) data18.2%
12Grok 4· Gaia2 (Meta ARE) data15.7%
13Qwen3 235B (Web)thinking· Gaia2 (Meta ARE) data15.7%
14GPT OSS 120Bhigh· Gaia2 (Meta ARE) data13.7%
15Qwen3 235B (Web)· Gaia2 (Meta ARE) data11.6%
16GPT 4o· Gaia2 (Meta ARE) data7.4%
17Llama 4 Maverick· Gaia2 (Meta ARE) data7.4%
18Llama 3.3 70B Instruct· Gaia2 (Meta ARE) data4.4%

Data credits: 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. · Data by LLM Stats