Recent releases from Anthropic, including Claude Opus 4.7 variants, have driven top Coding Arena scores on Arena.AI to the 1530–1535 Elo range through superior performance in blind human evaluations of code generation, refactoring, and debugging tasks. This positions frontier models like Claude ahead of OpenAI’s GPT-5.4 Codex series and Google’s Gemini 3.1 previews, which trail by 20–50 points amid ongoing scaling improvements in large language model training. With roughly 45 days until the June 30 resolution, traders are watching for potential new checkpoints or upgrades from major labs that could add 20–50 Elo points, though historical patterns show incremental rather than sudden jumps in arena leaderboards.