Anthropic’s Claude Fable 5 and Opus variants currently lead the Arena overall leaderboard with Elo ratings near 1505–1508, reflecting sustained human preference in blind evaluations across text, reasoning, and multi-turn tasks. Recent competitive pressure comes from OpenAI’s GPT-6 Astra family, which posted strong gains on specialized code arenas, and Google’s Gemini 3.8 Flash updates that lifted several category scores. Open-weight contenders such as Kimi K3 and Qwen3.8 Max trail by roughly 15–30 points, widening the closed-model advantage. With only weeks remaining until September 30, trader sentiment hinges on whether any lab can ship a capability jump—via new training runs, post-training enhancements, or router optimizations—sufficient to push the top overall score materially higher before the cutoff.