IBM and Kaggle launch continuous ScarfBench leaderboard for Java migration agents

IBM and Kaggle have launched a continuously updated public leaderboard for ScarfBench, IBM's open benchmark testing coding agents on enterprise Java migrations across Jakarta EE, Quarkus and Spring. Across 204 tasks, Gemini 4 Argon with the Antigravity harness leads the smoke metric at 75.2%, followed by Claude Opus 5.5 with Claude Code at 70.5%.

• 204 migration tasks and 10 model/harness configurations evaluated
• Gemini 4 Argon + Antigravity leads smoke metric at 75.2%, Claude Opus 5.5 + Claude Code at 70.5%
• On strict all-tests-pass, Claude Opus 5.5 with Claude Code tops at 72.1%
• About 72% of failing runs were runtime behavioral failures, not build errors