Vending-Bench 2
Running a simulated vending-machine business for a year.
As of 19 Sept 2026, Claude Opus 5 leads Vending-Bench 2 on BenchLeader with 11181.9, ahead of Claude Opus 4.7 at 10936.8, across 60 model configurations with a published result.
- Published by
- Andon Labsdata via Epoch AI Benchmarking Hub
- Category
- Agents & tools
- Index weight
- Reference only
- Models
- 60
- Data as of
- 19 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
The agent manages a vending business over a simulated year: ordering, pricing, and dealing with suppliers; scored on final balance.
How it is scored
Final balance in simulated dollars, as published by Andon Labs.
What to keep in mind
A dollar figure, not a percent; long-horizon coherence is what it tests.
- 1Claude Opus 511181.9
- 2Claude Opus 4.710936.8
- 3GPT-5.6 Sol9619.4
- 4Grok 4.69047.0
- 5GLM-5.28313.8
- 6GLM-5.38163.6
- 7Claude Opus 4.68017.6
- 8GPT-5.57523.8
- 9GPT-5.6 Terra7343.2
- 10Claude Sonnet 4.67204.1
- 11Muse Spark 1.16520.5
- 12Claude Sonnet 56377.7
- 13Kimi K2.66204.6
- 14GPT-5.46144.2
- 15GPT-5.3 Codex5940.1
60 of 60
| # | ||||
|---|---|---|---|---|
| 1 |  Claude Opus 5Anthropic | 11181.9 | 66.7 | 2026-07-24 |
| 2 | Claude Opus 4.7Anthropic | 10936.8 | 64.5 | 2026-04-16 |
| 3 |  GPT-5.6 SolOpenAI | 9619.4 | 56.8 | 2026-07-09 |
| 4 |  Grok 4.6xAI | 9047.0 | 59.3 | 2026-08-12 |
| 5 |  GLM-5.2Zhipu AIopen ↗ | 8313.8 | 52.1 | 2026-06-16 |
| 6 | GLM-5.3Zhipu AIopen ↗ | 8163.6 | – | 2026-08-14 |
| 7 | Claude Opus 4.6Anthropic | 8017.6 | 63.7 | 2026-02-05 |
| 8 | GPT-5.5OpenAI | 7523.8 | 63.2 | 2026-04-23 |
| 9 | GPT-5.6 TerraOpenAI | 7343.2 | – | 2026-07-09 |
| 10 | Claude Sonnet 4.6Anthropic | 7204.1 | 59.0 | 2026-02-17 |
| 11 |  Muse Spark 1.1Meta | 6520.5 | 65.2 | 2026-07-09 |
| 12 | Claude Sonnet 5Anthropic | 6377.7 | 57.2 | 2026-06-30 |
| 13 |  Kimi K2.6Moonshot AIopen ↗ | 6204.6 | 60.5 | 2026-04-20 |
| 14 | GPT-5.4OpenAI | 6144.2 | 59.2 | 2026-03-05 |
| 15 | GPT-5.3 CodexOpenAI | 5940.1 | 54.2 | 2026-02-05 |
| 16 | Claude Opus 4.8Anthropic | 5787.4 | 61.9 | 2026-05-28 |
| 17 | Claude Fable 5highAnthropic | 5680.3 | 64.0 | 2026-06-09 |
| 18 | GLM-5.1Zhipu AIopen ↗ | 5634.4 | 56.9 | 2026-04-07 |
| 19 |  Gemini 3 ProGoogle | 5478.2 | 61.1 | 2025-11-18 |
| 20 | Claude Fable 5.1Anthropic | 5421.6 | 64.5 | 2026-09-01 |
| 21 | Gemini 3.5 FlashGoogle | 5396.4 | 51.3 | 2026-05-19 |
| 22 | Kimi K3Moonshot AIopen ↗ | 5165.0 | 64.2 | 2026-07-16 |
| 23 |  Qwen3.6 PlusAlibabaopen | 5114.9 | 58.5 | 2026-03-31 |
| 24 | Kimi K2.7 CodeMoonshot AIopen ↗ | 5082.9 | 55.9 | 2026-06-12 |
| 25 | Claude Fable 5lowAnthropic | 5018.5 | – | 2026-06-09 |
| 26 | Claude Opus 4.5Anthropic | 4967.1 | 58.2 | 2025-11-24 |
| 27 | Claude Fable 5maxAnthropic | 4966.6 | 65.8 | 2026-06-09 |
| 28 | Grok 4.20xAI | 4662.9 | – | 2026-02-17 |
| 29 | Claude Fable 5Anthropic | 4529.9 | 68.3 | 2026-06-09 |
| 30 | GLM-5Zhipu AIopen | 4432.1 | 54.7 | 2026-02-11 |
| 31 | Claude Fable 5mediumAnthropic | 4339.8 | – | 2026-06-09 |
| 32 | Qwen3 6maxAlibaba | 4254.2 | 63.3 | 2026-04-20 |
| 33 | GPT-5.6 LunaOpenAI | 4094.7 | – | 2026-07-09 |
| 34 | Grok 4.5xAI | 3887.4 | 60.3 | 2026-07-08 |
| 35 | Claude Sonnet 4.5Anthropic | 3838.7 | 54.3 | 2025-09-29 |
| 36 | Gemini 3.1 Pro CustomtoolsGoogle | 3774.3 | – | 2026-02-19 |
| 37 | Gemini 3 FlashGoogle | 3634.7 | 57.7 | 2025-12-17 |
| 38 | GPT-5.2OpenAI | 3591.3 | 58.4 | 2025-12-11 |
| 39 |  DeepSeek V4 ProDeepSeekopen ↗ | 3284.5 | 52.6 | 2026-04-24 |
| 40 | Claude Opus 4.8maxAnthropic | 2992.3 | 64.2 | 2026-05-28 |
| 41 | GLM-4.7Zhipu AIopen ↗ | 2376.8 | 52.4 | 2025-12-22 |
| 42 |  MiniMax-M3MiniMaxopen ↗ | 2157.8 | 57.5 | 2026-06-01 |
| 43 | GPT-5.1OpenAI | 1473.4 | 57.2 | 2025-11-13 |
| 44 | Kimi K2.5Moonshot AIopen | 1198.5 | 53.9 | 2026-01-27 |
| 45 | Grok 4.1thinkingxAI | 1106.6 | 56.0 | 2025-11-19 |
| 46 | DeepSeek V3.2DeepSeekopen | 1034 | 56.0 | 2025-09-29 |
| 47 | Gemini 3.1 ProGoogle | 911.2 | 63.9 | 2026-02-19 |
| 48 | Gemini 2.5 ProGoogle | 573.6 | 54.5 | 2025-06-17 |
| 49 | Gemini 2.5 FlashGoogle | 548.8 | 52.5 | 2025-06-17 |
| 50 | Qwen3.5 FlashAlibaba | 462.7 | 52.5 | 2026-02-25 |
| 51 | Claude Haiku 4.5Anthropic | 458.9 | 48.1 | 2025-10-15 |
| 52 | Qwen3.5 27BAlibaba | 202.0 | 53.0 | 2026-02-24 |
| 53 | MiniMax-M2MiniMaxopen | 160.6 | 55.1 | 2025-10-27 |
| 54 | Qwen3 MaxmaxAlibaba | 71.6 | 53.5 | 2025-09-24 |
| 55 | Grok 4.3xAI | 35.3 | 50.5 | 2026-04-17 |
| 56 | Qwen3.5 PlusAlibaba | 0.5 | – | 2026-02-16 |
| 57 | Qwen3 235B A22B 2507thinkingAlibabaopen | -11.3 | 51.1 | 2025-07-25 |
| 58 | gpt-oss-120bOpenAIopen | -21.5 | 46.9 | 2025-08-05 |
| 59 | MiniMax-M2.5MiniMaxopen ↗ | -23.2 | 54.2 | 2026-02-12 |
| 60 | GPT-5 miniOpenAI | -31.2 | 56.8 | 2025-08-07 |
Cite as: BenchLeader, “Vending-Bench 2 leaderboard”, https://www.benchleader.com/benchmarks/vending_bench_2, data as of 19 Sept 2026.
Vending-Bench 2: questions
- What does Vending-Bench 2 measure?
- The agent manages a vending business over a simulated year: ordering, pricing, and dealing with suppliers; scored on final balance. Scores are reported in score; higher is better.
- Which AI model leads Vending-Bench 2?
- Claude Opus 5 leads Vending-Bench 2 with 11181.9 as of 19 Sept 2026, ahead of Claude Opus 4.7 at 10936.8.
- How many models have Vending-Bench 2 results?
- 60 model configurations have a Vending-Bench 2 result on BenchLeader, all taken from Andon Labs via Epoch AI Benchmarking Hub.
- Who runs Vending-Bench 2 and how often is it updated?
- Vending-Bench 2 is published by Andon Labs. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does Vending-Bench 2 count toward the BenchLeader Index?
- No. Vending-Bench 2 is shown for reference but left out of the composite index.