Benchmarks / Vending-Bench 2

Reported by Vending-Bench 2

Vending-Bench 2

Bank balance after an agent runs a simulated vending business for a year: ordering stock, setting prices and dealing with suppliers.

Last updated 1 Oct 2026

Results dated
1 Oct 2026
Results
63 configurations of 58 models
Unit
US dollars
Licence
Published with permission (Andon Labs results via Epoch AI)
Runs
Averaged over Andon Labs' runs; the count is not published

Vending-Bench 2 in brief

What it measures. Vending-Bench 2 is a third-party benchmark, shown here with credit. Measured here: Bank balance after an agent runs a simulated vending business for a year: ordering stock, setting prices and dealing with suppliers.

Latest results. , 63 configurations ranked.

Top 3 on Money after a year (US dollars, higher is better):

  1. GPT-6 Astra (OpenAI): $15,515
  2. GPT-6 Sol (OpenAI): $14,428
  3. Claude Opus 5 (Anthropic): $11,182

How to cite. Vending-Bench 2, via Spring Prompt, results dated 1 October 2026. https://springprompt.com/benchmarks/vending-bench-2 These results belong to Vending-Bench 2, not to us. Licence: Published with permission (Andon Labs results via Epoch AI). Credit Vending-Bench 2 first.

Money after a year: Muse Spark 1.1

Top 15 of 63 results · US dollars, higher is better. Choose a model to highlight it.Clear highlight

  1. 1 GPT-6 AstraOpenAI $15,515
  2. 2 GPT-6 SolOpenAI $14,428
  3. 3 Claude Opus 5Anthropic $11,182
  4. 4 Claude Opus 4.7Anthropic $10,937
  5. 5 GPT-5.6 SolOpenAI $9,619
  6. 6 Grok 4.6xAI $9,047
  7. 7 GLM-5.2Z.ai $8,314
  8. 8 GLM-5.3Z.ai $8,164
  9. 9 Claude Opus 4.6Anthropic $8,018
  10. 10 GPT-5.5OpenAI $7,524
  11. 11 GPT-5.6 TerraOpenAI $7,343
  12. 12 Claude Sonnet 4.6Anthropic $7,204
  13. 13 Muse Spark 1.1Meta $6,520
  14. 14 Claude Sonnet 5Anthropic $6,378
  15. 15 Kimi K2.6Moonshot AI $6,205

Full results

Vending-Bench 2: money after a year, US dollars, higher is better
#ModelMoney after a year
US dollars, higher is better
Price
$ per million tokens, in / out
1 GPT-6 AstraOpenAI
$15,515
$10 / $50
2 GPT-6 SolOpenAI
$14,428
$2 / $10
3 Claude Opus 5Anthropic
$11,182
$5 / $25
4 Claude Opus 4.7Anthropic
$10,937
$5 / $25
5 GPT-5.6 SolOpenAI
$9,619
$4 / $20
6 Grok 4.6xAI
$9,047
$2 / $6
7 GLM-5.2Z.ai
$8,314
$1.40 / $4.40
8 GLM-5.3Z.ai
$8,164
$1.40 / $4.40
9 Claude Opus 4.6Anthropic
$8,018
$5 / $25
10 GPT-5.5OpenAI
$7,524
$5 / $30
11 GPT-5.6 TerraOpenAI
$7,343
$2 / $12
12 Claude Sonnet 4.6Anthropic
$7,204
$3 / $15
13 Muse Spark 1.1Meta
$6,520
$1.25 / $4.25
14 Claude Sonnet 5Anthropic
$6,378
$2 / $10
15 Kimi K2.6Moonshot AI
$6,205
$0.95 / $4
16 GPT-5.4OpenAI
$6,144
$2.50 / $15
17 GPT-5.3-CodexOpenAI
$5,940
$1.75 / $14
18 Claude Opus 4.8Anthropic
$5,787
$5 / $25
19 Claude Fable 5 (high reasoning)Anthropic
$5,680
$10 / $50
20 GLM-5.1Z.ai
$5,634
$1.33 / $4.18
21 Gemini 3 Pro PreviewGoogle
$5,478
–
22 Claude Fable 5.1Anthropic
$5,422
$10 / $50
23 Gemini 3.5 FlashGoogle
$5,396
$1.50 / $9
24 Kimi K3Moonshot AI
$5,165
$3 / $15
25 Qwen3.6-PlusAlibaba
$5,115
$0.33 / $1.95
26 Gemini 3.8 FlashGoogle
$5,094
$1.50 / $7.50
27 Kimi K2.7 CodeMoonshot AI
$5,083
$0.95 / $4
28 Claude Fable 5 (low reasoning)Anthropic
$5,019
$10 / $50
29 Claude Opus 4.5Anthropic
$4,967
$5 / $25
29 Claude Fable 5 (max reasoning)Anthropic
$4,967
$10 / $50
31 Grok 4.20xAI
$4,663
$1.25 / $2.50
32 Claude Fable 5Anthropic
$4,530
$10 / $50
33 GLM-5Z.ai
$4,432
$1 / $3.20
34 Claude Fable 5 (medium reasoning)Anthropic
$4,340
$10 / $50
35 Qwen3.6-Max-PreviewAlibaba
$4,254
–
36 GPT-5.6 LunaOpenAI
$4,095
$0.20 / $1.20
37 Grok 4.5xAI
$3,887
$2 / $6
38 Claude Sonnet 4.5Anthropic
$3,839
$3 / $15
39 Gemini 3.1 Pro Preview (custom tools)Google
$3,774
$2 / $12
40 Gemini 3 Flash PreviewGoogle
$3,635
$0.50 / $3
41 GPT-5.2OpenAI
$3,591
$1.75 / $14
42 DeepSeek-V4-Pro (0423)DeepSeek
$3,285
$1.42 / $2.83
43 Claude Opus 4.8 (max reasoning)Anthropic
$2,992
$5 / $25
44 GLM-4.7Z.ai
$2,377
$0.60 / $2.20
45 MiniMax-M3MiniMax
$2,158
$0.30 / $1.20
46 GPT-5.1OpenAI
$1,473
$1.25 / $10
47 Kimi K2.5Moonshot AI
$1,198
$0.57 / $2.85
48 Grok 4.1 Fast (reasoning)xAI
$1,107
–
49 DeepSeek-V3.2-ExpDeepSeek
$1,034
$0.27 / $0.41
50 Gemini 3.1 Pro PreviewGoogle
$911.21
$2 / $12
51 Gemini 2.5 ProGoogle
$573.64
$1.25 / $10
52 Gemini 2.5 FlashGoogle
$548.84
$0.30 / $2.50
53 Qwen3.5-FlashAlibaba
$462.69
$0.065 / $0.26
54 Claude Haiku 4.5Anthropic
$458.89
$1 / $5
55 Qwen3.5-27BAlibaba
$201.98
$0.27 / $2.16
56 MiniMax-M2MiniMax
$160.60
–
57 Qwen3-MaxAlibaba
$71.57
–
58 Grok 4.3xAI
$35.26
$1.25 / $2.50
59 Qwen3.5-Plus (2026-02-15)Alibaba
$0.54
$0.26 / $1.56
60 Qwen3-235B-A22B-Thinking-2507Alibaba
$-11.34
$0.45 / $3.50
61 gpt-oss-120bOpenAI
$-21.53
$0.10 / $0.50
62 MiniMax-M2.5MiniMax
$-23.16
$0.29 / $1.20
63 GPT-5 miniOpenAI
$-31.18
$0.25 / $2

Swipe the table sideways for more columns.

Source: Vending-Bench 2, via Spring Prompt, springprompt.com/benchmarks/vending-bench-2. Test your own prompt on these models →

Ranks follow the score as shown, so equal numbers share a rank. Results as published by Vending-Bench 2; we do not re-run them.

Progress over time

Money after a year, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's; 2 models without a known release date are left off. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

$-5,000$0$5,000$10,000$15,000$20,000 Apr 2025JulOct2026AprJulOct Gemini 2.5 ProClaude Sonnet 4.5Gemini 3 Pro PreviewClaude Opus 4.6Claude Opus 4.7Claude Opus 5GPT-6 Astra

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 4 others not drawn.

OpenAIAnthropicxAIZ.aiMetaMoonshot AI

$-5,000$0$5,000$10,000$15,000$20,000 Apr 2025JulOct2026AprJulOct OpenAIAnthropicxAIZ.aiMetaMoonshot AI

What it measures

Bank balance after an agent runs a simulated vending business for a year: ordering stock, setting prices and dealing with suppliers.

What it does not measure

Not a real business; one simulated market with set rules.

Andon Labs; collected by Epoch AI. Licence: Published with permission (Andon Labs results via Epoch AI).