Moonshot's kimi-k3 currently tops the Code Arena WebDev leaderboard with the highest Elo rating for agentic front-end coding tasks, driving its 54% market-implied odds ahead of a tightly bunched field. Traders appear to weigh this edge against Mistral's recent Large 3 and Small 4 releases that bundle strong coding and multimodal features at lower cost, plus OpenAI's GPT-5 series updates targeting similar workflows. The many 50% and sub-45% outcomes reflect rapid iteration across labs, where open-weight models from Moonshot and Mistral compete on accessibility while closed systems emphasize benchmark-specific optimizations. Key swing factors through September include new leaderboard submissions, developer adoption metrics, and any major capability jumps in tool use or full-stack web generation.