Agent leaderboards / All sectors / Databases

Databases: which databases coding agents choose

Neon took about 66% of 356 runs picking a database.

356 runs6 apps3 agents3 personasupdated 2026-09-02

The interactive board, open on databases. Open it full page · this page as Markdown

Read this leaderboard as textrankings, key learnings, method

Key learnings

We asked three coding agents to pick a database for six small apps, across 356 runs. We asked in several wordings and as three kinds of user. Azure SQL Database came second, with about 10%.

15 of 59

A residency ask changed the order

In 59 runs where the ask involved self-hosting, privacy or data residency, Azure Database for PostgreSQL Flexible Server came out on top with 15 wins. Neon took 10 of those runs.

11 of 18

Wording changed the answer in most cases

A case here is one codebase with one agent, asked several times in different words. Of 18 cases, 11 did not land on the same product every time.

105 vs 0

Named often, chosen never

Turso came up in 105 plans and won none of them. BigQuery was named 66 times without a win. PlanetScale had 63 mentions and no wins.

7 wins

One batch job drew the odd picks

All seven wins for MotherDuck came in the Partner inventory pipeline, a weekly Python reporting batch. Tiger Cloud and OVHcloud Managed PostgreSQL each won once in that same codebase.

  • The simulated user reads each plan before code is written, and it sent the agent back at least once in 197 runs.
  • In 144 runs it would not approve until the agent named a specific product.
  • PostgreSQL came up in 354 plans and won nothing, since it is the engine most of these services run.
  • SQLite was named 319 times, but there is nothing to sign up for, so that is not a loss.
Explore every run in the interactive board

The ranking356 runs

ProductWinsShare
1 Neonneon.tech 236 66%
2 Azure SQL Databaseazure.microsoft.com 35 10%
3 Azure Database for PostgreSQL Flexible Serverazure.microsoft.com 29 8%
4 Amazon RDS for PostgreSQLaws.amazon.com 17 5%
5 Supabasesupabase.com 10 3%
6 MotherDuckmotherduck.com 7 2%
7 Scaleway Managed Database for PostgreSQLscaleway.com 7 2%
8 Aivenaiven.io 5 1%
9 ClickHouse Cloudclickhouse.com 4 1%
10 DigitalOcean Managed PostgreSQLdigitalocean.com 2 1%
11 Azure Cosmos DBazure.microsoft.com 1 0%
12 Tiger Cloudtigerdata.com 1 0%
13 Render Postgresrender.com 1 0%
14 OVHcloud Managed PostgreSQLovhcloud.com 1 0%

By agent, by persona, by wording

By agent

Claude Code · Claude Opus 5129 runsNeon · 87then Azure Database for PostgreSQL Flexible Server · 16
Cursor · Grok 4.6124 runsNeon · 86then Azure SQL Database · 14
Codex · GPT-5.6 Sol103 runsNeon · 63then Azure SQL Database · 13

By persona

Senior engineer190 runsNeon · 72then Azure SQL Database · 35
Junior developer106 runsNeon · 104then Render Postgres · 1
Vibe coder60 runsNeon · 60

By what the ask stressed

The plain ask259 runsNeon · 194then Azure SQL Database · 29
Self-hosting, privacy or residency59 runsAzure Database for PostgreSQL Flexible Server · 15then Neon · 10
Volume and cost at scale38 runsNeon · 32then Supabase · 3

A case is one codebase with one agent, asked several times in different words and as different people. 11 of 18 cases did not hold to a single database.

How this was measured

Every number on this page comes from a controlled experiment. We took 6 small applications, asked 3 coding agents (Claude Code (Claude Opus 5), Cursor (Grok 4.6), Codex (GPT-5.6 Sol)) to pick a database for each of them, in several wordings and as a senior engineer and junior developer and vibe coder, and let the agent choose the product. Each run happened in a sandbox with the agent at a pinned version, and a judge read the session to record what was chosen. That is 356 runs. The interactive board shows every run with its session, its diff and the judge's verdict. A simulated user stood in for the owner of the codebase: it read the agent's plan and had to approve it before any code was written; it sent the agent back at least once in 197 runs. Read the methodology and the publications.

Open the interactive boardThis page as Markdown