智谱2月12日还发布GLM Coding Plan价格调整函，因市场需求增长，用户规模与调用量提升，公司加大 算力与模型优化投入，产品能力持续升级；公司决定对套餐价格体系进行结构性调整，取消首购优惠， 保留按季按年订阅优惠，套餐价格整体涨幅自30%起，已订阅用户价格保持不变。智谱官方GLM Coding Plan上线即售罄。国产AI编程模型的付费套餐被抢空，在行业中较为罕见。 智谱（02513）早盘一度涨超20%，本周累计涨幅已近140%。截至发稿，涨18.16%，报475港元，成交 额8.63亿港元，总市值2118亿港元。 2月12日，智谱AI正式上线并开源其新一代旗舰模型GLM-5。智谱官方确认，这个新模型正是此前在全 球模型服务平台OpenRouter登顶热度榜首、引发圈内疯狂猜测的神秘模型"Pony Alpha"。 另外值得注意的是，继H股上市后，智谱继续推进其A股上市计划。近日，中国证监会官网更新了智谱 公司的IPO辅导进展信息。根据证监会披露，智谱撤回了其于2025年4月提交的辅导备案，并办理了新 的辅导备案登记。与上次不同的是，智谱IPO辅导机构变更为两家券商，由原来的中金公司变更为国泰 海通证 ...

Core Insights - The launch of Doubao-Seed-Code, the first programming model from the Doubao model family by ByteDance's Volcano Engine, focuses on optimizing Agentic Coding tasks and offers a competitive price-performance ratio [1][3][33] Performance and Features - Doubao-Seed-Code outperforms several domestic models like DeepSeek-V3.1, Kimi-K2, and GLM-4.6, with scores only second to the top model Claude Sonnet 4.5, and features a native context of 256K, surpassing Claude Sonnet 4.5's 200K [1][3] - The model supports visual understanding, allowing it to generate code from UI design drafts, screenshots, or hand-drawn sketches, significantly enhancing front-end development efficiency [3][19] - Doubao-Seed-Code integrates seamlessly with popular development tools, enabling users to switch from Claude Code with minimal learning curve [7][31] Cost Efficiency - The model employs a tiered pricing model, with input costs at 1.20 yuan per million tokens and output costs at 8.00 yuan per million tokens, achieving a 62.7% reduction in overall usage costs with full transparent caching [4][31] Real-World Application - Doubao-Seed-Code has demonstrated capabilities in real programming scenarios, such as autonomously planning development tasks, quickly building front-end web pages, and modifying databases while actively fixing errors and optimizing structures [6][16] - The model can create functional prototypes based on detailed prompts, showcasing its ability to handle complex development tasks effectively [17][25] Training and Development - The model was trained using a large-scale Agent reinforcement learning system, utilizing a dataset covering 100,000 container images and providing an end-to-end sandbox environment for evaluation [27][29] - Doubao-Seed-Code's training process emphasizes pure reinforcement learning, achieving state-of-the-art performance in software engineering tasks without the need for distilled or labeled cold-start data [29] Market Position - The emergence of Doubao-Seed-Code addresses the supply risks faced by overseas AI programming models, providing developers with a stable and controllable alternative [33]

Core Insights - Alibaba has launched a new open-source AI programming model, Qwen3-Coder, which has achieved top programming capabilities in the global open-source model landscape, surpassing closed-source models like GPT-4.1 and rivaling the strongest programming model, Claude4 [1] Group 1 - The Qwen3-Coder model has made significant breakthroughs in code generation and agent invocation capabilities [1] - With Qwen3-Coder, entry-level programmers can accomplish in one day what experienced programmers would typically complete in a week [1] - The model enables the creation of a brand website in as little as five minutes [1]

Core Viewpoint - Deepseek-R1 0528 demonstrates strong performance in front-end development tasks, comparable to OpenAI's Opus 4 and surpassing Sonnet 4 and Gemini 2.5 Pro, especially considering the price difference [3][4][51]. Group 1: Model Performance Comparison - In front-end capabilities, Deepseek-R1 0528 slightly lags behind Opus 4 but outperforms Sonnet 4 and Gemini 2.5 Pro [3]. - Deepseek-R1 0528 successfully completed complex tasks that Opus 4 struggled with, although the quality and completion rate were slightly lower [3][4]. - The price of Deepseek-R1 0528 is significantly lower than Opus 4, making its performance even more impressive [4][51]. Group 2: Testing Results - In the warehouse management system test, Deepseek-R1 0528 produced a professional interface with complete functionality, while other models failed to deliver usable outputs [11]. - For the dot animation editor, Deepseek-R1 0528 excelled, providing a fully functional interface, while other models either failed to animate or had significant issues [17]. - In the gradient color extraction tool test, Deepseek-R1 0528 showcased excellent aesthetic design but failed to implement the color extraction logic, while Opus 4 and Sonnet 4 managed to complete the functionality albeit with simpler designs [20][21]. Group 3: Overall Implications - The advancements in Deepseek-R1 0528 suggest a shift in the AI programming model landscape, where high-quality outputs can be achieved at a fraction of the cost compared to leading competitors [51]. - The performance of Deepseek-R1 0528 indicates a potential democratization of access to advanced AI tools, allowing more users to leverage powerful models without prohibitive costs [51].