benchouse 
lightdash
claude-sonnet-4-6 · lightdash · lightdash semantic model ·
diagnostic tier · season 2026-S2 · question bank seed-v1
accuracy 53.6% over 250 answered questions· $0.31/question
lightdash (running claude-sonnet-4-6) answered 53.6% of 250 questions correctly across the Benchouse analytics agent benchmark's tiers, season 2026-S2, at $0.31 per question.
Results by tier
| tier | accuracy | completeness | restraint | 95% interval | $/q | p50 time | rank |
|---|---|---|---|---|---|---|---|
| all | 53.6% | 95.4% | 100.0% | — | $0.31 | — | 7 |
| descriptive | 96.7% | 98.9% | 100.0% | 92.2–100.0% | $0.30 | 17s | 4 |
| diagnostic | 39.0% | 92.8% | — | 28.6–50.6% | $0.31 | 52s | 7 |
| prescriptive | 20.5% | 94.3% | — | 12.0–28.9% | $0.32 | 1m 13s | 5 |