The Mid-Autumn Festival and National Day holidays are poised to become the most bustling time in the AI industry this year, as AI companies from both home and abroad are set to concentratedly unveil new models. In Silicon Valley, Anthropic's Claude Opus 5.2 has commenced grayscale testing, demonstrating significant enhancements in performance. Google is rumored to be testing Gemini 4 Pro under an anonymous guise, with its coding, agent, reasoning, and other capabilities reportedly outperforming those of its competitors. Elon Musk has announced that Grok 4.7 is nearing the end of its training phase and will integrate SpaceX's proprietary data to bolster its engineering application prowess.

On the domestic front, JieYue XingChen has released Step 5 Preview, which boasts a sparse Mixture of Experts (MoE) architecture and supports a context window of up to 1 million tokens. It excels in software engineering scenarios and offers a single-task cost that is merely one-eighth of that of Claude Opus 5. This model will be open-sourced on October 15. Kimi K3.1 is anticipated to be launched in September. DeepSeek's V4.1 Flash model is also on the horizon, with performance expected to surpass that of the existing V4 Pro. Alibaba has open-sourced the preview model Qwen3.8-Flash-Next, based on the Qwen4 architecture, as it prepares for the full-scale launch of the Qwen4 model family.