Skip to content

Benchmark

Vending-Bench 2

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

Models measured
69
Scores recorded
87
Sources
3
IQ weight
2.11

Score distribution

6,546.7
Mean
5,165
Median
8,742.6
Std dev
458.9
10th percentile
11,181.9
90th percentile
-31.2
Min
-31.280,017.6
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.11 for chat models, measured over 69 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51Claude Fable 5· Epoch AI Benchmarking Hub data4,529.9
52GLM 5· Epoch AI Benchmarking Hub data4,432.1
53GLM 5GLM-5 official release table · average net assets· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.4,432.1
54Claude Fable 5medium· Epoch AI Benchmarking Hub data4,339.8
55Qwen3.6 Max Preview· Epoch AI Benchmarking Hub data4,254.2
56GPT-5.6 Luna· Epoch AI Benchmarking Hub data4,094.7
57Grok 4.5· Epoch AI Benchmarking Hub data3,887.4
58Claude Sonnet 4.5final balance · launch-page chart· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.3,849.7
59Claude Sonnet 4.5· Epoch AI Benchmarking Hub data3,838.7
60Claude Sonnet 4.5final balance · system-card prose· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.3,838.7
61Gemini 3.1 Pro Previewcustomtools· Epoch AI Benchmarking Hub data3,774.3
62Gemini 3 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
3,635
63Gemini 3 Flashmean final net worth· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.3,635
64Claude Haiku 5.5· Epoch AI Benchmarking Hub data3,602.5
65GPT 5.2· Epoch AI Benchmarking Hub data3,591.3
66DeepSeek V4 Pro· Epoch AI Benchmarking Hub data3,284.5
67Claude Opus 4.8max· Epoch AI Benchmarking Hub data2,992.3
68GLM 4.7· Epoch AI Benchmarking Hub data2,376.8
69MiniMax M3· Epoch AI Benchmarking Hub data2,157.8
70GPT 5.1· Epoch AI Benchmarking Hub data1,473.4
71Kimi K2.5· Epoch AI Benchmarking Hub data1,198.5
72Grok 4.1 Fastreasoning· Epoch AI Benchmarking Hub data1,106.6
73DeepSeek V3.2 Exp· Epoch AI Benchmarking Hub data1,034
74Gemini 3.1 Pro Preview· Epoch AI Benchmarking Hub data911.2
75Gemini 2.5 Pro· Epoch AI Benchmarking Hub data573.6
76Gemini 2.5 Flash· Epoch AI Benchmarking Hub data548.8
77Qwen3.5 Flash· Epoch AI Benchmarking Hub data462.7
78Claude Haiku 4.5· Epoch AI Benchmarking Hub data458.9
79Qwen3.5 27B· Epoch AI Benchmarking Hub data202
80MiniMax M2· Epoch AI Benchmarking Hub data160.6
81Qwen3 Max· Epoch AI Benchmarking Hub data71.6
82Grok 4.3· Epoch AI Benchmarking Hub data35.3
83Qwen3.5 Plus· Epoch AI Benchmarking Hub data0.544
84Qwen 3 235B A22B Thinking 2507· Epoch AI Benchmarking Hub data-11.3
85GPT OSS 120B· Epoch AI Benchmarking Hub data-21.5
86MiniMax M2.5· Epoch AI Benchmarking Hub data-23.2
87GPT-5 mini· Epoch AI Benchmarking Hub data-31.2

Showing 51–87 of 87 results.

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (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. · Data by LLM Stats