How it works

AntQL · 7 October 2026 · Updated 23 October

Top score on BIRD LiveSQLBench

We achieved the top score on all three BIRD LiveSQLBench leaderboards: SQLite, Base and Large. We use the benchmark to evaluate and compare data agents.

The top five of the BIRD LiveSQLBench SQLite leaderboard. 1: AntQL 1.1, AntHill AI, 72.22. 2: Sage Agent V2, SpyneLabs, 71.11. 3: Agent Sentinel V1, Genloop, 68.15. 4: AntQL 1.0, AntHill AI, 62.96. 5: Sage Agent, SpyneLabs, 57.41.
The top five of the BIRD LiveSQLBench SQLite leaderboard at livesqlbench.ai, captured 7 October 2026, 08:59 IST (03:29 UTC).
The top five of the BIRD LiveSQLBench Base leaderboard (Base-Full v1). 1: AntQL 1.1 (Claude Opus 5.5), AntHill AI, 57.00. 2: AskRUI, Dipeak, 50.33. 3: DIA (Data Intelligence Agents), C3 AI, 48.00. 4: Jitto Build (Claude Opus 5.5, high), JamLabs, 41.33. 5: MiniMax M3 (Claude Code), MiniMax, 40.17.
The top five of the BIRD LiveSQLBench Base leaderboard (Base-Full v1) at livesqlbench.ai, captured 23 October 2026, 12:04 IST (06:34 UTC).
The top five of the BIRD LiveSQLBench Large leaderboard (Large-v1). 1: AntQL 1.1 (Claude Opus 5.5), AntHill AI, 53.33. 2: GPT-5.5 (xhigh) (Codex CLI), OpenAI, 35.21. 3: GPT-5.5 (low), OpenAI, 33.12. 4: GPT-5.5 (low) (OpenHands CLI), OpenAI, 32.50. 5: GPT-5.5 (low) (Codex CLI), OpenAI, 32.50.
The top five of the BIRD LiveSQLBench Large leaderboard (Large-v1) at livesqlbench.ai, captured 23 October 2026, 12:04 IST (06:34 UTC).

Benchmarks offer useful direction, but they can’t fully capture the complexity of real-world use. We use them as one way to measure progress.

Answering questions accurately is one part of what we’re building. Anthill is a holistic system that learns the context of your organisation: your data, your definitions and how your teams work. With that context, ant. works proactively, spotting problems and moving work forward before anyone asks, with a person approving anything consequential.