ClickHouse Releases Second Edition of Agentic Analytics Benchmark

ClickHouse published the second round of its Agentic Analytics Benchmark on October 20, 2026, adding five frontier models to an existing, fixed leaderboard, with outside coverage of the release following on November 12, 2026. The five additions are GPT-6 Astra, GPT-6 Luna, GPT-6 Sol, Claude Opus 5.5 and Claude Sonnet 5.5. The benchmark is a public test run by ClickHouse in which participating models answer 201 analytics questions against a synthetic data warehouse, each writing its own SQL inside an agentic loop, and all models are evaluated on the same questions, the same warehouse and a shared judge panel. The board now holds 34 models in total.

In the second round, the top of the leaderboard barely moved. Claude Fable 5.1 kept the lead with a 76.6% win rate, while the five newly added entries landed between rank 9 and lower positions. Rather than relying on a single headline score, the benchmark tracks finer metrics such as the number of rounds taken, truncation rate and cost, which are meant to show how models differ on complex, multi-step analytics work. With the second round now out, the leaderboard and its rankings rest at this round's results, and the benchmark has not yet been updated further.