API

据澎湃新闻报道，阿里巴巴集团 CEO 吴泳铭在全体会议上发表演讲，称 AI 推动千行百业生产力变革，将创造远超互联网链接所创造的价值，“面向 AI 时代，阿里巴巴专注做好两件事，一是不断完善 AI 基础设施，二是坚持开源开放，把 AI 能力转化为千行百业的生产力。” 吴泳铭进一步透露，过去一年，阿里巴巴在 AI 基础设施方面进行了高强度投入。得益于规模效应和技术进步，算力成本持续降低。其中，阿里大…

快手旗下的可灵 AI 宣布，API 能力免费升级，号称“加量不加价”： 支持 1080P 高清视频输出的 V1.5 高品质模型，和可灵 V1.0 模型相比，画面质量、动态质量、文本响应度等方面显著提升。 开放 V1.0 模型的视频延长功能，对可灵 AI 生成后的视频可续写 4～5 秒，支持多次续写（最长 3 分钟），可通过微调提示词进行视频续写创作，实现更丝滑的场景转换。 可灵 AI 官方表示，A…

In a Nov. 8 blog post, tech media outlet WinBuzzer reported that Mistral AI has launched a new content review API to meet the growing demand for a secure online environment. Citing a press release, the content vetting API is based on a fine-tuned Ministral 8B model that categorizes content into nine categories, including hate speech, violence, and personal data breaches. The content review API supports Chinese, Arabic, English, French, German, Italian, Japanese, Korean, Portuguese, Russian and Spanish...

Elon Musk's artificial intelligence company xAI has officially released its application programming interface (API), which allows developers to integrate Grok into other applications. Musk made the announcement on Monday, posting a link to the interface on the xAI website. Previously, Musk had confirmed plans to provide an API for Grok in August and open-sourced the software's weights in March. To view and use the API, users must log into their xAI account and select "Request Ac...

Recently, Smart Spectrum AI announced that the video call API:GLM-4-Plus-VideoCall went live on the open platform for open applications. According to the introduction, this is the model behind the clear language video call, capable of realizing cross-modal capabilities such as video call and voice multi-round interaction, which is expected to become a new paradigm to promote the innovation of human-computer interaction. Previously, the latest base big model GLM-4-Plus developed by Beijing Zhi Spectrum Huazhang Technology Co., Ltd. was officially launched and has been deployed on its open platform bigmodel.cn. GLM-4-Plus in the code computation, data analysis, image and video feature recognition...

Recently, Shutterbug AI has ushered in another big upgrade and officially opened its API service. This upgrade update is colorful and powerful, including video generation, image generation, virtual try-on, etc. Users can easily place self-service orders without a tedious application process. One of the new features is "Counterpart". This feature is very interesting, users can upload their own voiceover or singing on the character video generated by Kerin AI, and then the system will automatically synchronize the audio with the mouth shape of the video character perfectly. The videos generated by Kerin 1.0 model and Kerin 1.5 model, as long as they meet the video screen...

September 29th, 2012, Wisdom AI released the latest promotion for October this afternoon, with a minimum discount of 10% on all official model APIs of the open platform bigmodel.cn, and 100 million token credits for each user during the period. In addition, during the month of October, users can enjoy step discounts on the following day based on their token usage on the previous day. Daily consumption of more than 100 million tokens, the next day all API calls 50% discount Daily consumption of more than 1 billion tokens, the next day all API calls 30% discount ...

Runway, a New York-based AI startup and a well-known name in the field of AI-generated video technology, has launched an API for its advanced video models.This will enable developers and companies to integrate Runway's generative AI capabilities into their own applications and services. The API, which currently has limited access, offers Runway's Gen-3 Alpha Turbo model - a faster variant of its flagship Gen-3 Alpha product. Pricing starts at 1 cent per credit for generative...

Recently, venture capitalist Vinod Khosla has become a hot topic for a controversial statement he made on social media. He claimed that "100% of AI unicorns are built on OpenAI's API". This statement immediately triggered strong opposition and doubts from netizens, and many people expressed dissatisfaction with it. Netizens commented that Khosla's statement was a bit exaggerated. One user said that "70% of future AI unicorns are turning to Claude", while others directly refuted:…

Beijing Zhipu Huazhang Technology Co., Ltd. recently announced that it will open the API interface of its GLM-4-Flash large language model to the public for free to promote the popularization and application of large model technology. The GLM-4-Flash model has shown significant advantages in both speed and performance, especially in terms of inference speed. By adopting optimization measures such as adaptive weight quantization, parallel processing technology, batch processing strategy and speculative sampling, it has achieved a stable speed of up to 72.14 tokens/s, which is outstanding among similar models. In terms of performance optimization, GLM-4-Fl…

OpenAI has introduced a new feature, Structured Outputs, which aims to make the output generated by the model strictly follow the JSON schema provided by the developer. At last year's DevDay, OpenAI first introduced JSON Schema, which provides an important foundation for developers to build reliable applications. Although JSON Schema improves the reliability of models generating valid JSON outputs, it does not ensure that the model's response conforms to a specific schema. Today's Structured Outputs feature…

Yi API, a subsidiary of 01AI, recently announced the launch of the Yi-Large-FC model with Function Call. This new feature enables the model to determine when to call external tools based on user input and respond in JSON format, which is compatible with OpenAI interface design and can smoothly replace GPT. Function Call capability is essential for building AI Agent systems, especially in scenarios that require real-time information, dedicated task processing, automated workflows, and enhanced model capabilities. Yi-Large-…

iFlytek announced that the long-context version of iFlytek Spark API, Spark Pro-128K large model, is officially available, with the lowest price of 0.21 yuan/10,000 tokens. According to reports, the conversation between users and large models is usually considered to be short-term memory. Once the length of the conversation exceeds its context carrying capacity, the excess part may be forgotten by the model. Different from traditional text processing models, long text models have more accurate text understanding and generation capabilities and stronger cross-domain migration capabilities. They can understand and generate more information at one time, suitable for...

In June this year, OpenAI announced that it would stop providing API services to China. Some industry insiders analyzed that this would hit some companies that used OpenAI’s large model to start a business. Recently, an investor asked iFlytek on an interactive platform: OpenAI shut down its API services in China. Has iFlytek Mars taken over the API business decisions of OpenAI’s domestic users? In response, iFlytek replied: After OpenAI shut down its API services in China, the number of API calls to iFlytek Spark has increased significantly. This further verifies the autonomous and controllable universal large model base…

Yesterday, API developers took to social media to report that they had received a "warning letter" from OpenAI stating that they would be taking additional steps to stop API use in regions they do not support. The letter reads, "According to data, your organization is receiving API traffic from regions that OpenAl does not currently support. Starting July 9, we will be taking additional steps to stop API usage from countries and regions that are not on OpenAI's list of supported regions." Affected organizations that wish to continue using OpenAI's services...

Yesterday, API developers successively said on social media that they had received a "warning letter" from OpenAI, which stated that it would take additional measures to stop the use of APIs in regions it does not support. Since then, several domestic large model companies have announced that they will provide developers with alternatives to OpenAI API services, including Zhipu, Alibaba Cloud, Baidu Smart Cloud, etc. Zhipu: Launched a "special relocation plan" for OpenAI API users to help users switch to domestic large models. Its GLM large model claims to be fully comparable to the OpenAI product system,...

Security company Sysdig recently released a report claiming that a large number of hackers are targeting major LLM large language model network platforms to launch "LLMjacking" attacks. Hackers steal user accounts and passwords through a series of methods, resell model APIs to third parties, and select private information from user conversation records for blackmail or public sale. Sysdig said that hackers seem to "prefer" Anthropic Claude v2/v3 platforms. Currently, they have detected that hackers mainly use database collision and PHP frameworks...

Programmer Q&A platform Stack Overflow and OpenAI today announced a new API partnership. OpenAI and Stack Overflow will provide OpenAI users with the accurate, audited data foundation they need through the Overflow API so that AI tools can quickly find solutions to problems, allowing technicians to focus on high-priority tasks; OpenAI will also display questions from Stack Overflow directly in ChatGPT…

Moonshot AI announced that Kimi Big Model has learned to use tools, that is, the API has supported the Tool Calling feature. Through this feature, developers can let Kimi Big Model interact with various custom external tools, thereby expanding the imagination of AI applications. For example, Kimi Big Model can call map tools to display company addresses, call PPT production tools to make presentations, or call mathematical calculation tools to solve complex problems in conversations. Kimi Big Model will intelligently…

Baidu announced at today's AI Developer Conference that the number of users of Wenxin Yiyin has exceeded 200 million, while the average daily API calls have reached 200 million, which has roughly doubled since the company's update last December. Robin Li introduced, "Wenxin YiYin was released from March 16 last year, and today is one year and one month. Our number of users has exceeded 200 million, the average daily API calls have also exceeded 200 million, the number of customers served has reached 85,000, and the number of AI native apps developed using the Thousand Sails platform has exceeded 190,000." In addition, Wenxin Big Model 4.0 tool version was officially released, and the algorithm training efficiency was increased...

OpenAI launched the Batch batch processing API for developers this morning, which can give results within 24 hours and provide a 50% discount on the API. The new Batch API is suitable for asynchronous task processing. For example, when developers need to process a large amount of text, pictures, and summaries, they can use this API, and OpenAI will give the processing results within 24 hours. In this way, OpenAI can process during off-peak hours, save server resources, and provide developers with a 50% discount to unlock higher rate limits. The new Batch API…

After a long wait, OpenAI has finally announced the full opening of the GPT-4 API, a news that, like the announcement of a technological revolution, brings new opportunities for the development of generative AI applications.GPT-4 not only makes breakthroughs in terms of creativity, but also possesses stronger contextualization capabilities, which greatly expands its application scenarios. GPT-4 has achieved significant advances in creativity, not only generating content, but also editing and iterating ideas with users, providing powerful support for tasks such as music composition and script writing. Its powerful creativity allows users to easily engage with GPT-4 in...

OpenAI recently released a press release announcing improvements to the fine-tuning API and further expansion of the custom model plan. The press release contains the following improvements to the fine-tuning API: Epoch-based Checkpoint Creation. During each training epoch (a training process that passes all samples in the training dataset once (and only once)), a complete fine-tuning model checkpoint is automatically generated to reduce the need for subsequent retraining, especially in the case of overfitting…

The good old days of developers whoring out Gemini are over! Recently, some developers reported receiving an email from the Gemini API team informing them that Google will be updating the additional terms of service for the Gemini API starting May 2, 2024, and will begin charging for API requests made through cloud billing accounts. The change primarily affects users of the Gemini API provided by Google AI for Developers (known as the "Generative Language API" in the Cloud Console), and does not affect Vertex AI G...