JetBrains publishes a 105-task Kotlin benchmark for coding agents

Published on 8 July 2026, the Kotlin Benchmark scores agents on 105 engineering tasks drawn from eight active open-source Kotlin repositories, following the SWE-bench pattern: the agent gets a real issue description, and a task counts as resolved only when the generated patch passes the required tests inside a containerised environment. In the first public run, Claude Code with Opus 4.7 at xhigh effort resolved 90 of 105 tasks (85.71%), ahead of JetBrains Junie with Opus 4.7 max and Codex with GPT-5.5 xhigh, both at 81.9%. The dataset and test harnesses are public on GitHub and the leaderboard is open.

What it means Nearly every public agent benchmark is Python. A language-specific, test-verified suite is the only honest way to know whether a headline score transfers to the stack you actually ship.