# Visa Agentic Commerce Benchmark leaderboard

Whether an AI shopping assistant finds out what a customer needs, then picks a matching product or declines when nothing fits.

## Visa Agentic Commerce Benchmark: pass rate % (higher is better, 22 models)

| Rank | Model | Provider | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Argon | Google | 82.2 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Muse Wonder | Meta | 82.2 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Claude Fable 5.1 | Anthropic | 75.6 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | GLM 5.3 | Zhipu AI | 66.7 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Claude Sonnet 5.5 | Anthropic | 64.4 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | GPT-6 Astra | OpenAI | 64.4 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Muse Spark 1.3 | Meta | 64.4 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Claude Opus 5.5 | Anthropic | 62.2 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | GPT-6.1 Sol | OpenAI | 62.2 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Grok 4.7 | xAI | 60 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Qwen 3.8 Max | Qwen | 60 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | DeepSeek V4.1 Flash | DeepSeek | 57.8 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 1 | Beam 1.0 | Reflection AI | 55.6 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | Gemini 3.8 Flash | Google | 53.3 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | GPT-6 Luna | OpenAI | 53.3 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | Kimi K3.1 Pro | Moonshot AI | 53.3 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | GPT-6.1 Terra | OpenAI | 51.1 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | MiniMax M3 | MiniMax | 48.9 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | DeepSeek V4 Pro | DeepSeek | 42.2 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | Gemini 3.1 Pro | Google | 35.6 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | MiniMax M3.1 Flash | MiniMax | 33.3 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |
| 14 | Inkling | Thinking Machines | 28.9 | pass rate % | 15 private episodes, 3 runs each | Official board | Visa Research, Oct 2026 snapshot | https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark | 5 Nov 2026 (checked 6 Nov 2026) |

Scores are never combined across benchmarks. Each row names its metric, source and date.

## About this benchmark

- Maker: Visa (United States)
- Year: 2026
- Use case: Financial advice and safety
- Size: 15 private episodes, 3 runs each; 22 models in the first snapshot
- Public split: none: episodes are private; one can be played as a demo
- Grading: A simulated customer reveals requirements only when asked. A run passes if the assistant finds every requirement and the right buy or no-buy decision is made, checked against a hidden answer key.
- Licence: Not stated
- Submission: None published; Visa invites model developers' input
- Products allowed: Models only, in Visa's harness
- Latest version: October 2026 snapshot (released 2026-11-05)
- What it does not cover: Checkout, payment and order placement (planned).
- Last checked: 6 Nov 2026
- Links: paper https://www.visapaymentsfrontier.io/articles/agentic-commerce-benchmark/assets/visa-agentic-commerce-benchmark-whitepaper.pdf, site https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark, leaderboard https://www.visapaymentsfrontier.io/#agentic-commerce-benchmark
- Directory page: https://benchmarks.finance/benchmarks/visa-agentic-commerce/
- Leaderboard: https://benchmarks.finance/leaderboards/visa-agentic-commerce/

Source: https://benchmarks.finance/leaderboards/visa-agentic-commerce/

Created and maintained by Moneda.