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