BIRD
Harbor: pendingText-to-SQL parsing, which aims at converting natural language instructions into executable SQLs, has gained increasing attention in recent years.
Incomplete14%Coding
- Samples
- N/A
- Released
- 2023
- Scores
- 0
AptMQL-Bench
AptMQL-Bench: From Text-to-SQL to Text-to-MQL via Access-Pattern Schema Design and Data-Preserving Migration
A text-to-MQL benchmark with 21 document databases and 3,186 natural language requests, converted from BIRD using access-pattern schema design.
How to run
ltb · aptmql-bench
$ npm i -g lettokenburn-cli$ ltb bench info aptmql-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Opus 4.5 (2025-11-01) | 70.3 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Claude Sonnet 4.5 (2025-09-29) | 67.9 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Claude Haiku 4.5 | 64.7 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| GPT-oss 120B | 62.5 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| google/gemma-4-31B-it | 62.0 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Qwen-3.5-35B | 56.1 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| gpt-oss-20b | 51.0 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Llama-3.1-70B-Instruct | 37.1 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
AptMQL-Bench comprises 21 document-oriented databases and 3,186 natural language requests with associated MQL queries. It is constructed by converting the BIRD text-to-SQL benchmark using a pipeline that designs schemas from access patterns and migrates data without loss.
Scoring uses soft execution accuracy. The strongest model, Claude Opus 4.5, achieves 70.34% accuracy with external knowledge, indicating that realistic text-to-MQL generation remains challenging.