By analyzing 16,900 sessions, we can see what tools each of Claude Code, Codex, and Cursor chose.
2026-09-07 14:37Models🔥 40.2 heat score
1sources
1days unfolding
40.2heat score
4mentions
SummaryAI generated
Armature Company conducted direct task tests on three mainstream programming agents: Claude Code, Codex, and Cursor, using 16,893 session data (including 1,163 variant prompts, 75 code repositories, and 10 programming languages). After filtering, 5,292 valid sessions were obtained, covering 51 code libraries and 18 industry sectors.
Based on 16,893 sessions data (including 1,163 prompt variants, 75 code repositories, and 10 languages), Armature company conducted tests on three mainstream coding agents: Claude Code, Codex, and Cursor. In total, 5,292 valid sessions were identified, covering 51 code repositories and 18 industries.