E-Commerce Bench: leaderboard
Eighteen agents are handed ¥100,000 each and told to run up to four simulated online stores for 365 days against real market data — sourcing, negotiating with suppliers, pricing, managing inventory and staying solvent. The score is end-of-year total assets averaged over five independent episodes, which is why the board spans three orders of magnitude and why two models went bankrupt in 2 of their 5 runs. Qwen, 2026.
Metric: Final Assets (¥k, mean of 5 episodes). Source: github.com. Saturation forecast: Around October 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 1431 | #19 (GPT-5.6 Sol) |
| 2 | Claude Fable 5 (Max) | 805 | #16 (Claude Fable 5) |
| 3 | GPT-5.5 | 702 | #31 |
| 4 | Claude Opus 4.8 (Max) | 498 | #37 (Claude Opus 4.8) |
| 5 | Qwen 3.8 Max | 472 | #35 |
| 6 | GLM-5.3 (Max) | 391 | #39 (GLM-5.3) |
| 7 | GLM-5.2 (High) | 301 | #73 (GLM-5.2) |
| 8 | Kimi K3 | 265 | #30 |
| 9 | Claude Opus 4.7 (Max) | 259 | #49 (Claude Opus 4.7) |
| 10 | GLM-5.3 Flash (Max) | 258 | #65 (GLM-5.3 Flash) |
| 11 | Claude Opus 4.6 (Max) | 258 | #63 (Claude Opus 4.6) |
| 12 | GLM-5.1 | 226 | #97 |
| 13 | Gemini 3.5 Flash | 190 | #60 |
| 14 | Qwen 3.7 Max | 165 | #75 |
| 15 | Gemini 3.1 Pro (Preview) | 130 | #58 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=e-commerce-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.