The Independent Analytics Agent Benchmark
Every configuration answers the same 300 questions (100 descriptive, 100 diagnostic, 100 prescriptive) against a deliberately messy simulated warehouse with withheld ground truth. No vendor can buy a position, or removal. How it is measured → Compare any two →
| 1 | Snowflake Cortex Agents  claude-opus-4-8 · snowflake semantic view | 83.7% | 99.7% | 100.0% | $0.59 | 1 |
| 2 |  Hex  gpt-5.6 (Hex) · hex semantic model (dbt) | 74.4% | 100.0% | 100.0% | $0.35 | 2 |
| 3 |  nao claude-sonnet-5 · nao table docs | 70.0% | 100.0% | 100.0% | $0.95 | 3 |
| 4 |  Supersimple LLM not disclosed · supersimple semantic model | 69.7% | 98.3% | 100.0% | – | 4 |
| 5 | Databricks Genie LLM not disclosed · genie space instructions | 65.7% | 68.0% | 80.0% | – | 5 |
| 6 |  Snowflake Cortex Analyst claude-sonnet-4-6 · snowflake semantic view | 54.8% | 81.0% | 100.0% | $0.13 | 6 |
| 7 |  Lightdash AI claude-sonnet-4-6 · lightdash semantic model | 53.6% | 95.4% | 100.0% | $0.31 | 7 |
Season 2026-S2 · question bank seed-v1 · universe seed 42 · judge claude-opus-4-8-r4.