FrontierCode

coding agentscodereasoning

Modality: text · Language: en · Multilingual: no · Max score: 1

FrontierCode is Cognition's coding evaluation that tests whether models can pass difficult coding tasks while meeting the standards of high-quality production codebases. The Diamond subset contains the hardest problems.

Methodology

Imported from llm-stats public benchmark metadata. Modality: text. Max score: 1. Categories: agents, code, reasoning. Language: en.