Recent releases from OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 and Mythos 5.1 in early September have kept Anthropic models atop the Arena.ai text leaderboard near 1505–1508 Elo, with Google’s Gemini 3.8 Flash and Meta’s Muse Spark variants close behind. Traders are watching whether these or follow-on updates can push any model past common resolution thresholds like 1510–1530 before September 30, as incremental gains in agentic coding, reasoning benchmarks, and human preference votes have narrowed gaps among the frontier labs. Chinese open-weight efforts from Alibaba and Moonshot remain a few points behind but add competitive pressure. With just weeks remaining, the next credible capability jump or fine-tune from the top three providers will likely determine whether the current ceiling holds or moves higher.