Vending Bench 2 benchmark scores from Epoch AI's benchmark data collection.
Sourced from Epoch AI's systematic benchmark data collection covering frontier models.
Auto-generated definition — editorial description not yet written.
Higher is better.
Auto-generated definition — editorial description not yet written.
| # | Model | Score |
|---|---|---|
| 1 | Kimi K2.6 | 6204.6 |
| 2 | GPT-5.3 Codex (xhigh) | 5940.1 |
| 3 | Claude Opus 4.7 | 5826.7 |
| 4 | GLM-5.1 (Reasoning) | 5634.4 |
| 5 | Qwen3.6 Plus | 5114.9 |
| 6 | Kimi K2.7 Code | 5082.9 |
| 7 | Grok 4.20 0309 v2 (Reasoning) | 4662.8 |
| 8 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | 4529.9 |
| 9 | GLM-5 (Reasoning) | 4432.1 |
| 10 | Qwen3.6 Max Preview | 4254.2 |
| 11 | Google: Gemini 3 Flash Preview | 3436.1 |
| 12 | GLM-4.7 (Reasoning) | 2376.8 |
| 13 | MiniMax-M3 | 2157.8 |
| 14 | Anthropic: Claude Sonnet 4.5 | 2125.1 |
| 15 | Anthropic: Claude Haiku 4.5 | 1700.5 |
| 16 | Gemini 3 Pro Preview (low) | 1235.2 |
| 17 | Kimi K2.5 (Reasoning) | 1198.5 |
| 18 | Claude Opus 4.5 (Non-reasoning) | 1040.3 |
| 19 | Gemini 3.1 Pro Preview | 911.2 |
| 20 | Gemini 2.5 Pro | 772.5 |
| 21 | Grok 4.1 Fast (Reasoning) | 723.1 |
| 22 | Qwen: Qwen3.5-Flash | 462.7 |
| 23 | Gemini 2.5 Flash (Non-reasoning) | 280.8 |
| 24 | Qwen3.5 27B (Reasoning) | 202.0 |
| 25 | MiniMax-M2 | 160.6 |
| 26 | Qwen3 Max | 71.6 |
| 27 | Grok 4.3 (high) | 35.3 |
| 28 | Qwen: Qwen3.5 Plus 2026-02-15 | 0.5 |
| 29 | Qwen3 235B A22B 2507 (Reasoning) | -11.3 |
| 30 | gpt-oss-120b (high) | -21.5 |
| 31 | MiniMax-M2.5 | -23.2 |