Skip to content

Benchmark

BALROG MiniHack

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

Models measured
37
Scores recorded
38
Sources
1
IQ weight
1.48

Score distribution

16.3%
Mean
10%
Median
14.8%
Std dev
2.5%
10th percentile
37.5%
90th percentile
0%
Min
0%65%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.48 for chat models, measured over 37 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT-6 Astramax· Epoch AI Benchmarking Hub data65%
2GPT-5.6 Solmax· Epoch AI Benchmarking Hub data42.5%
3Gemini 3 Pro Preview· Epoch AI Benchmarking Hub data40%
4Claude Opus 5max· Epoch AI Benchmarking Hub data37.5%
5GPT-5.6 Terramax· Epoch AI Benchmarking Hub data37.5%
6Gemini 3.1 Pro Preview· Epoch AI Benchmarking Hub data35%
7Claude Opus 4.5· Epoch AI Benchmarking Hub data30%
8Claude Opus 4.564k· Epoch AI Benchmarking Hub data30%
9Gemini 3 Flash· Epoch AI Benchmarking Hub data30%
10DeepSeek R1· Epoch AI Benchmarking Hub data25%
11GPT-5.6 Lunamax· Epoch AI Benchmarking Hub data22.5%
12Grok 3beta· Epoch AI Benchmarking Hub data22.5%
13GPT 5minimal· Epoch AI Benchmarking Hub data20%
14Gemini 2.5 Pro· Epoch AI Benchmarking Hub data17.5%
15Grok 4· Epoch AI Benchmarking Hub data17.5%
16Claude 3.5 Sonnet· Epoch AI Benchmarking Hub data15%
17Claude Haiku 4.5· Epoch AI Benchmarking Hub data15%
18Gemini 2.5 Flash· Epoch AI Benchmarking Hub data12.5%
19Claude Haiku 3.5· Epoch AI Benchmarking Hub data10%
20GPT 4o· Epoch AI Benchmarking Hub data10%
21GPT-4o mini· Epoch AI Benchmarking Hub data10%
22Reka Flash 3· Epoch AI Benchmarking Hub data10%
23Llama 3.1 70B Instruct· Epoch AI Benchmarking Hub data7.5%
24gemini-1.5-flash-002· Epoch AI Benchmarking Hub data5%
25gemini-1.5-pro-002· Epoch AI Benchmarking Hub data5%
26Llama 3.1 8B Instruct· Epoch AI Benchmarking Hub data5%
27Llama 3.2 1B Instruct· Epoch AI Benchmarking Hub data5%
28Llama 3.2 90B Vision Instruct· Epoch AI Benchmarking Hub data5%
29Llama 3.3 70B Instruct· Epoch AI Benchmarking Hub data5%
30Phi 4· Epoch AI Benchmarking Hub data5%
31Qwen2.5 72B Instruct· Epoch AI Benchmarking Hub data5%
32DeepSeek R1 Distill Qwen 32B· Epoch AI Benchmarking Hub data2.5%
33Llama 3.2 11B Vision Instruct· Epoch AI Benchmarking Hub data2.5%
34Llama 3.2 3B· Epoch AI Benchmarking Hub data2.5%
35Mistral Nemo· Epoch AI Benchmarking Hub data2.5%
36Qwen2 VL 7b Instruct· Epoch AI Benchmarking Hub data2.5%
37Qwen2-VL (72B) Instruct· Epoch AI Benchmarking Hub data2.5%
38Qwen2.5 7B Instruct· Epoch AI Benchmarking Hub data0%

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence)