Recent advances in agentic coding have driven strong trader interest in whether leading large language models will hit elevated scores on benchmarks like SWE-bench Verified or Coding Arena by June 30. Anthropic’s Claude Mythos Preview currently tops verified leaderboards near 94 percent through refined tool use and multi-step reasoning, while OpenAI’s GPT-5.3 Codex and Google’s Gemini 3 series sit in the mid-80s to low-90s range on comparable tasks. Competitive pressure from frequent updates, including high-reasoning modes and specialized coding agents released earlier this year, continues to push incremental gains. Traders watch for any new model drops or benchmark refreshes in the coming weeks, as historical patterns show rapid iteration can close remaining gaps before the deadline, though saturation effects on established evals introduce uncertainty.