SShogo Bench

Vending-Bench 2

Run a simulated vending business for a year starting with $500. Score is the final bank balance (log-scaled).

Agent Index weight
6%
in the Long-horizon autonomy pillar
Models tested
68
from the benchmark’s leaderboard and reports
Normalised for the index
log scale from 500 to 20000
Official site

Results

#ModelScoreNormalisedHarnessCost/taskDate
1Gemini 4 Carbon Supported
Google
$16,484source——
2GPT-6 Astra Supported
OpenAI
$15,515source——
3Gemini 4 Hydrogen Estimated
Google
$14,549source——
4GPT-6 Sol Estimated
OpenAI
$14,428source——
5Gemini 4 Argon Supported
Google
$13,718source——
6GPT-6.1 Sol Estimated
OpenAI
$12,861source——
7Claude Opus 5 Supported
Anthropic
$11,182source——
8Claude Sonnet 5.5 Estimated
Anthropic
$11,165source——
9Claude Opus 4.7 Supported
Anthropic
$10,937source——
10Grok 4.7 Estimated
xAI
$10,537source——
11Claude Fable 5.2 Supported
Anthropic
$9,628source——
12GPT-5.6 Sol Supported
OpenAI
$9,619source——
13Muse Wonder Estimated
Meta
$9,432source——
14Claude Opus 5.5 Supported
Anthropic
$9,235source——
15GPT-6.1 Terra Estimated
OpenAI
$9,107source——
16Grok 4.6 Supported
xAI
$9,047source——
17GLM-5.2 Supported
Z.ai (Zhipu)
$8,314source——
18GLM-5.3 Supported
Z.ai (Zhipu)
$8,164source——
19Claude Opus 4.6 Supported
Anthropic
$8,018source——
20GPT-5.5 Supported
OpenAI
$7,524source——
21GPT-5.6 Terra Estimated
OpenAI
$7,343source——
22Claude Sonnet 4.6 Supported
Anthropic
$7,204source——
23Muse Spark 1.1 Estimated
Meta
$6,520source——
24Claude Sonnet 5 Supported
Anthropic
$6,378source——
25Kimi K2.6 Estimated
Moonshot AI
$6,205source——
26GPT-5.4 Supported
OpenAI
$6,144source——
27GPT-5.3 Codex Supported
OpenAI
$5,940source——
28Claude Opus 4.8 Supported
Anthropic
$5,787source——
29Claude Fable 5 Supported
Anthropic
$5,680source——
30GLM-5.1 Estimated
Z.ai (Zhipu)
$5,634source——
31Gemini 3 Pro Preview Supported
Google
$5,478source——
32Claude Fable 5.1 Supported
Anthropic
$5,422source——
33Gemini 3.5 Flash Supported
Google
$5,396source——
34Kimi K3 Supported
Moonshot AI
$5,165source——
35Qwen3.6 Plus Estimated
Alibaba (Qwen)
$5,115source——
36Gemini 3.8 Flash Supported
Google
$5,094source——
37Kimi K2.7 Code Estimated
Moonshot AI
$5,083source——
38Claude Opus 4.5 (latest) Supported
Anthropic
$4,967source——
39Grok 4.20 (Non-Reasoning) Estimated
xAI
$4,663source——
40GLM-5 Supported
Z.ai (Zhipu)
$4,432source——
41Qwen3.6 Max Preview Estimated
Alibaba (Qwen)
$4,254source——
42GPT-5.6 Luna Estimated
OpenAI
$4,095source——
43Grok 4.5 Estimated
xAI
$3,887source——
44Claude Sonnet 4.5 (latest) Supported
Anthropic
$3,839source——
45Gemini 3.1 Pro Preview Supported
Google
$3,774source——
46Gemini 3 Flash Preview Supported
Google
$3,635source——
47Claude Haiku 5.5 Estimated
Anthropic
$3,602source——
48GPT-5.2 Supported
OpenAI
$3,591source——
49DeepSeek V4 Pro Estimated
DeepSeek
$3,285source——
50GLM-4.7 Estimated
Z.ai (Zhipu)
$2,377source——
51MiniMax-M3 Estimated
MiniMax
$2,158source——
52GPT-5.1 Supported
OpenAI
$1,473source——
53Kimi K2.5 Supported
Moonshot AI
$1,198source——
54Grok 4.1 Fast (Non-Reasoning)
xAI
$1,107source——
55DeepSeek V3.2 Supported
DeepSeek
$1,034source——
56Gemini 2.5 Pro Supported
Google
$574source——
57Gemini 2.5 Flash Estimated
Google
$549source——
58Claude Haiku 4.5 Estimated
Anthropic
$459source——
59GPT-5 Mini Estimated
OpenAI
$-31source——
60GPT OSS 120B Supported
OpenAI
$-22source——
61Grok 4.3 Beta
xAI
$35source——
62MiniMax-M2 Estimated
MiniMax
$161source——
63MiniMax-M2.5 Estimated
MiniMax
$-23source——
64Qwen3 235B A22B Thinking 2507
Alibaba (Qwen)
$-11source——
65Qwen3.5 27B
Alibaba (Qwen)
$202source——
66Qwen3.5 Flash
Alibaba (Qwen)
$463source——
67Qwen3.5 Plus Estimated
Alibaba (Qwen)
$1source——
68Qwen3 Max Estimated
Alibaba (Qwen)
$72source——

For models without a Vending-Bench 2 result, the index estimates one from Epoch’s Capabilities Index using a logistic fit on 62 models (residual SD 1.34 on the logit scale). Estimates are labelled and widen the model’s range.