EnterpriseMem-Bench: leaderboard
EnterpriseMem-Bench (Tummalapenta and Addanki, 2026) is a multi-turn Text-to-SQL benchmark of 300 analyst sessions (1,400 turns) over real BIRD Financial, SEC EDGAR and Northwind databases, in which follow-up questions refer back to entities and years from earlier turns. Execution accuracy in percent: a turn is correct when the generated SQL returns the ground-truth result set; higher is better. This board uses the paper's Condition B, a two-turn working-memory window plus resolved context keys.
Metric: Execution accuracy (%; all 1,400 turns of 300 multi-turn Text-to-SQL sessions over BIRD Financial, SEC EDGAR and Northwind; Condition B: two-turn working memory plus resolved context keys; single run). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 86.4 | #172 |
| 2 | GPT-5 Mini (2025-08-07) | 83.2 | #341 |
| 3 | Claude Sonnet 4.5 (Non-reasoning) | 79.8 | #244 (Claude Sonnet 4.5) |
| 4 | Claude Sonnet 4.6 (Non-reasoning) | 59.6 | #158 (Claude Sonnet 4.6) |
| 5 | Claude Opus 4.6 (Non-reasoning) | 58.7 | #100 (Claude Opus 4.6) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=enterprisemem-bench · How It Works · Data refreshed daily, snapshot 2026-11-12.