Enter opens the highlighted result. Esc closes.

Vending-Bench 2 leaderboard

GPT-6 Astra leads
$15,515
models measured
60
lead over #2
+$1086.85
median
$5,088

Leader on Vending-Bench 2 · read October 22

On October 22, 2026, 60 of our catalog's models had a Vending-Bench 2 result, and OpenAI GPT-6 Astra led with $15,515, ahead of OpenAI GPT-6 Sol at $14,428.

An agent runs a simulated vending machine business over a long stretch of simulated time: stocking, pricing, emailing suppliers. The number is the money it ends with, in US dollars, as Andon Labs runs it and Epoch AI collects it. One harness for every model.

Role in the index

Vending-Bench 2 is shown as published and does not feed any of our indexes. Read the methodology

Vending-Bench 2 results for models in the catalog
# Model Result Evidence Coding index Output $/Mtok
1 GPT-6 AstraOpenAI $15,515 85.1 $50.00
2 GPT-6 SolOpenAI $14,428 76.0 $10.00
3 Gemini 4 ArgonGoogle $13,718 81.0 $10.00
4 Claude Opus 5Anthropic $11,182 83.0 $25.00
5 Claude Sonnet 5.5Anthropic $11,165 87.3 $10.00
6 Claude Opus 4.7Anthropic $10,937 67.5 $25.00
7 Grok 4.7xAI $10,537 66.7 $6.00
8 GPT-5.6 SolOpenAI $9,619 73.5 $20.00
9 Muse WonderMeta $9,432 71.9 $20.00
10 Claude Opus 5.5Anthropic $9,235 88.8 $20.00
11 Grok 4.6xAI $9,047 65.7 $6.00
12 GLM 5.2Z.ai $8,314 61.1 $4.40
13 GLM 5.3Z.ai $8,164 70.6 $4.40
14 Claude Opus 4.6Anthropic $8,018 57.2 $25.00
15 GPT-5.5OpenAI $7,524 71.3 $30.00
16 GPT-5.6 TerraOpenAI $7,343 63.9 $12.00
17 Claude Sonnet 4.6Anthropic $7,204 57.3 $15.00
18 Muse Spark 1.1Meta $6,520 70.3 $4.25
19 Claude Sonnet 5Anthropic $6,378 57.7 $10.00
20 Kimi K2.6Moonshot AI $6,205 52.1 $4.00
21 GPT-5.4OpenAI $6,144 64.4 $15.00
22 GPT-5.3-CodexOpenAI $5,940 60.0 $14.00
23 Claude Opus 4.8Anthropic $5,787 65.0 $25.00
24 Claude Fable 5Anthropic $5,680 79.3 $50.00
25 GLM 5.1Z.ai $5,634 49.2 $4.40
26 Claude Fable 5.1Anthropic $5,422 84.2 $50.00
27 Gemini 3.5 FlashGoogle $5,396 39.8 $9.00
28 Kimi K3Moonshot AI $5,165 62.3 $15.00
29 Qwen3.6 PlusQwen $5,115 44.1 $1.95
30 Gemini 3.8 FlashGoogle $5,094 61.6 $3.75
31 Kimi K2.7 CodeMoonshot AI $5,083 53.4 $4.00
32 Claude Opus 4.5Anthropic $4,967 44.0 $25.00
33 Grok 4.20xAI $4,663 33.3 $2.50
34 GLM 5Z.ai $4,432 41.7 $3.20
35 GPT-5.6 LunaOpenAI $4,095 58.0 $1.20
36 Grok 4.5xAI $3,887 58.6 $6.00
37 Claude Sonnet 4.5Anthropic $3,839 39.5 $15.00
38 Gemini 3.1 Pro Preview Custom ToolsGoogle $3,774 Not benchmarked $12.00
39 Gemini 3 Flash PreviewGoogle $3,635 43.7 $3.00
40 Claude Haiku 5.5Anthropic $3,602 70.7 $0.50
41 GPT-5.2OpenAI $3,591 55.9 $14.00
42 DeepSeek V4 Pro 0423DeepSeek $3,285 38.7 $2.976
43 GLM 4.7Z.ai $2,377 55.1 $2.20
44 MiniMax M3MiniMax $2,158 39.2 $1.20
45 GPT-5.1OpenAI $1,473 40.4 $10.00
46 Kimi K2.5Moonshot AI $1,198 38.6 $2.85
47 DeepSeek V3.2DeepSeek $1,034 30.0 $0.88
48 Gemini 3.1 Pro PreviewGoogle $911 32.3 $12.00
49 Gemini 2.5 ProGoogle $574 19.1 $10.00
50 Gemini 2.5 FlashGoogle $549 Not benchmarked $2.50
51 Qwen3.5-35B-A3BQwen $463 Not benchmarked $1.30
52 Claude Haiku 4.5Anthropic $459 35.6 $5.00
53 Qwen3.5-27BQwen $202 Not benchmarked $1.56
54 MiniMax M2MiniMax $161 Not benchmarked $1.02
55 Grok 4.3xAI $35 37.4 $2.50
56 Qwen3.5 Plus 2026-04-20Qwen $1 39.5 $1.80
57 Qwen3 235B A22B Thinking 2507Qwen -$11 Not benchmarked $3.50
58 gpt-oss-120bOpenAI -$22 Not benchmarked $0.50
59 MiniMax M2.5MiniMax -$23 37.2 $1.20
60 GPT-5 MiniOpenAI -$31 34.2 $2.00

Only models in our catalog are listed, and each appears once with its best published run. 215 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out.