Benchouse Analytics Agent Benchmark - Prescriptive: leaderboard
Vendor analytics agents query a deliberately messy simulated e-commerce warehouse using their own semantic layers and default LLMs. Descriptive questions ask what happened, diagnostic questions ask why, and prescriptive questions ask what action to take. LLM-judged conditional accuracy counts correct answers among answered questions, not all 300 nominal questions; completeness and restraint are separate measures. Current season/question bank is consistent across three tiers of 100 nominal questions. Agent/model and semantic-layer configuration are retained; undisclosed model products remain omitted by existing policy.
Metric: Correct / answered questions (%). Source: benchouse.ai. Saturation forecast: Around August 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 55.17 | #22 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=benchouse-analytics-agent-benchmark-prescriptive · How It Works · Data refreshed daily, snapshot 2026-11-05.