altimate-code scores 71.7% Pass@1 on DAB, ahead of agents running Claude Opus 4.6.
About DAB
DAB (Data Agent Benchmark) is the first benchmark for evaluating data agents on realistic, complex, data-oriented tasks — a collaboration between the EPIC Data Lab at UC Berkeley and Hasura PromptQL. Unlike prior SQL-only or single-database benchmarks, DAB stresses agents under real enterprise data complexity: multi-database integration, ill-formatted key joins, unstructured-text transformation, and domain knowledge. It spans 54 queries across 12 datasets, 9 domains, and 4 database systems — PostgreSQL, MongoDB, SQLite, and DuckDB.
Test Configuration
Leaderboard
Official DAB leaderboard, Pass@1, ranked, as published June 2026. altimate-code scores 0.717.
Benchmark Composition
12 datasets spanning 4 database systems — every task requires the agent to work across heterogeneous stores.
| Dataset | Databases | DBs | Tables | Queries |
|---|---|---|---|---|
| agnews | MongoDB, SQLite | 2 | 3 | 4 |
| bookreview | PostgreSQL, SQLite | 2 | 2 | 3 |
| crmarenapro | DuckDB, PostgreSQL, SQLite | 6 | 27 | 13 |
| deps_dev_v1 | DuckDB, SQLite | 2 | 3 | 2 |
| github_repos | DuckDB, SQLite | 2 | 6 | 4 |
| googlelocal | PostgreSQL, SQLite | 2 | 2 | 4 |
| music_brainz_20k | DuckDB, SQLite | 2 | 2 | 3 |
| pancancer_atlas | DuckDB, PostgreSQL | 2 | 3 | 3 |
| patents | PostgreSQL, SQLite | 2 | 2 | 3 |
| stockindex | DuckDB, SQLite | 2 | 2 | 3 |
| stockmarket | DuckDB, SQLite | 2 | 2754 | 5 |
| yelp | DuckDB, MongoDB | 2 | 5 | 7 |
| Total | PostgreSQL · MongoDB · SQLite · DuckDB | 28 | 2,811 | 54 |
Sources
- DAB Leaderboard (UC Berkeley EPIC Data Lab) — Official leaderboard and benchmark overview
- DataAgentBench on GitHub — Benchmark repository, datasets, and submission guide
- altimate-code submission (PR #44) — Our leaderboard submission
- DAB paper (arXiv) — Methodology and analysis
DAB Bench FAQs
DAB (Data Agent Benchmark), published as DataAgentBench, is the first benchmark for evaluating data agents on realistic, complex, multi-database tasks — a collaboration between the EPIC Data Lab at UC Berkeley and Hasura PromptQL. Unlike SQL-only or single-database benchmarks, DAB tests agents on real enterprise data complexity: multi-database integration, ill-formatted key joins, unstructured-text transformation, and domain knowledge. It spans 54 queries across 12 datasets, 9 domains, and 4 database systems — PostgreSQL, MongoDB, SQLite, and DuckDB.
Altimate Code scored 71.7% Pass@1 on DAB Bench using GPT-5.5 + Claude Sonnet 4.6, ahead of Spacedock/Recce (67.2% on Opus 4.8), MinusX (65.2%), and Pi Coding Agent (61.0% on Opus 4.6). Altimate Code's submission is PR #53 on the DataAgentBench repository.
Altimate Code on GPT-5.5 + Sonnet 4.6 beats competitors running Opus 4.8 — a more capable, more expensive model. The reason is the harness: purpose-built, deterministic tools for data engineering outperform raw model capability. DAB requires multi-database reasoning across PostgreSQL, MongoDB, SQLite, and DuckDB simultaneously — exactly the kind of task that specialized tooling handles better than general-purpose code generation.
ADE-Bench (by dbt Labs) focuses on analytics engineering tasks using dbt projects on Snowflake and DuckDB. DAB Bench (by UC Berkeley EPIC Data Lab) tests agents on heterogeneous multi-database tasks — requiring agents to query and reason across PostgreSQL, MongoDB, SQLite, and DuckDB simultaneously. DAB spans 54 queries across 12 datasets and 9 domains, stressing agents on real enterprise data complexity rather than single-warehouse analytics work.
Yes — DAB Bench is open source on GitHub at github.com/ucbepic/DataAgentBench. The official leaderboard is at ucbepic.github.io/DataAgentBench. Altimate Code's submission is PR #44 on the DataAgentBench repository. The benchmark methodology is also published in a paper on arXiv (arxiv.org/abs/2603.20576).
DAB Bench covers 4 database systems: PostgreSQL, MongoDB, SQLite, and DuckDB — spanning 12 datasets and 28 individual databases across 54 queries. Every task requires the agent to work across heterogeneous stores simultaneously. Altimate Code also supports Snowflake, BigQuery, Databricks, Redshift, MySQL, and additional warehouses in production.
Reproduce This on Your Warehouse. Same Agent, Your Data.
The Altimate Code build behind these DAB results runs on your own dbt project. Install it, or book a demo and we run it with you.
