According to CryptoWorld, Redis founder Salvatore Sanfilippo refuted claims that large-scale Chinese AI models rely on distilling American models. He pointed out that ordinary APIs can only return textual answers, failing to capture the probability distribution and internal state of the model during answer generation. Therefore, it's impossible to replicate the core capabilities of cutting-edge models through a few external calls. He emphasized that the logical reasoning and analytical abilities of large-scale models are hidden within complex neural networks; external users can only obtain the final textual answer, not the complete thought process and probability calculations. Furthermore, regarding Washington and Anthropics' attempts to cut off the distillation channels for large-scale Chinese AI models by blocking cutting-edge models, former GitHub internationalization strategy head Kevin S. Xu pointed out that such blocking policies cannot hinder the overall progress of AI in China.

According to Bijie.com, Cursor previewed several major products at its inaugural Compile conference, including its self-developed 1.5 trillion parameter large-scale model and Origin, a Git-compatible intelligent agent native code hosting platform. SpaceX announced its acquisition of Cursor's parent company, Anysphere, for $60 billion. Cursor is transforming from an IDE-assisted tool into an AI-native software engineering platform, focusing on its general-purpose foundation model. This model is pre-trained from scratch on a Colossus cluster consisting of over 100,000 GPUs, aiming to surpass conventional code generation and possess intelligence such as independent planning, tool invocation, and software testing. The model is expected to be released in the coming weeks. The Origin platform is designed for large-scale intelligent agent collaboration and is planned to launch in the fall of 2026. Meanwhile, the Cursor Mobile app (iOS TestFlight) also began testing during the summit, supporting remote connection to Macs for AI programming.

According to AiYi's monitoring, an entity associated with a16z withdrew 88,350 HYPE tokens from various exchanges in the past half hour, worth approximately $6.4 million. Since 2026, this entity appears to have accumulated a position of 7.335 million HYPE tokens, totaling approximately $340 million, at an average price of $46.46, currently showing a floating profit of approximately $193 million.

According to a recent report by Anthropic, published by Odaily Planet Daily, the importance of programming skills themselves has decreased with the assistance of Claude Code, while management and domain knowledge have become key factors influencing outcomes. Claude Code users average 20 hours of activity per week. In ten major professions, the success rate of non-programmers using agents to write code has narrowed to less than 7% compared to professional software engineers, with management positions showing a slightly higher success rate than professional programmers. Users lead approximately 70% of planning direction decisions, while agents handle approximately 80% of specific execution decisions. Expert-level sessions can trigger 12 actions and 3200 words of output per session, more than five times that of novice-level sessions. Novice sessions have a 19% abandonment rate after encountering errors, while intermediate and expert users have abandonment rates of 5% to 7% when encountering problems. Within seven months, the proportion of debugging sessions used to fix broken code decreased from 33% to 19%, while the proportion of sessions for direct deployment, data analysis, and writing non-code documentation doubled, and the average estimated value of tasks increased by approximately 25%.

According to Bijie.com, Zhipu AI's GLM-5.2 achieved a score of 51 in the AI Large Model Intelligence Index V4.1 evaluation, surpassing Minimax-M3 (44 points), DeepSeek V4 Pro (44 points), and Kimi K2.6 (43 points) to become the world's top open-source model. In the GDPval-AA V2 test, GLM-5.2 scored 1524 (human benchmark score 1000), leading Minimax-M3 (1418) and DeepSeek V4 Pro (1328), and on par with GPT-5.5. Compared to GLM-5.1, scientific inference Critpt increased to 21%, and HLE increased to 40%. GLM-5.2's average cost per task is approximately $0.46, still lower than closed-source models in the same intelligence tier. It has a total of 744 bytes of parameters, 40 bytes of activation parameters, and the context window has increased from 200K to 1M. It is open-source under the MIT license. The official API of Zhipu is now online, with a pricing of 1.4 per input and 4.4 per million output tokens.