NEEDLE and Ouroboros at Practical ML Conf: Search Benchmark and...

NEEDLE and Ouroboros at Practical ML Conf: Search Benchmark and Agent That Outperformed Codex

At the Practical ML Conf (Yandex) conference on September 19, the author of the 'Love. Death. Transformers.' channel presents the open search engine benchmark NEEDLE, while the author of AbstractDL presents the coding agent Ouroboros, which outperformed Codex, Claude Code, and Cursor on open benchmarks.

Compiled by Sergey KostenchukPublished 2026-09-19Updated 2026-09-20

2026-09-19 Coding Anthropic

💻 NEEDLE and Ouroboros at Practical ML Conf

The author of 'Love. Death. Transformers.' is speaking today at Practical ML Conf (Yandex) about NEEDLE — an open benchmark for search engines ("Set" hall, 15:30–16:05). Anton from AbstractDL (@abstractDL) is currently breaking down Ouroboros — a coding agent with a constitution and memory that outperformed Codex, Claude Code, and Cursor.

🌍 NEEDLE fills a gap: benchmarks test either language models or search algorithms, but not the final engine (API) used by the agent. Static queries allow caching "correct" documents and tuning ranking — NEEDLE takes fresh queries from news, financial documents, and court cases, invalidating tuning.

👤 The "Set" hall is being broadcast live on YouTube — watch for free. From the talk — a material on choosing a search engine for agents.

Source 1: https://pmlconf.yandex.ru/2026/program Source 2: https://www.youtube.com/watch?v=8IWLBsQU7qE

Sources