News & Updates
July 16, 2026
Kimi K3 released: 2.8 trillion parameters, native multi-modality, million token context
Dark Side of the Moon released Kimi K3 on July 16: 2.8 trillion parameters, native multi-modal, 1M token context, specially built for long-range programming, knowledge work and deep reasoning; ARR has exceeded 200 million US dollars, K3 has consolidated its leading position in domestic long context models.
Source: Moonshot AI Official
Grok 4.5 released: Musk’s first large model after merging into SpaceX, covering all terminals in one week
xAI released Grok 4.5 on July 16. It is officially called "SpaceXAI's most intelligent model" and is specially designed for coding, agentic tasks and knowledge work. Within a week after the release, it has launched all grok.com/X/iOS/Android, and entered office and development ecosystems such as Google Workspace, Outlook, Excel and GitHub Copilot.
Source: xAI (SpaceXAI) official news
July 27, 2026
Dark Side of the Moon open source Kimi K3: 2.8 trillion parameters are fully open, overseas developers call it "local version of Fable 5"
Dark Side of the Moon open sourced Kimi K3 on July 27, the world's first open 3T-level parameter model (2.8 trillion), using MoE architecture, native visual capabilities, and 1 million token context; it topped the Hugging Face trend list as soon as it went online, and was called the "local version of Fable 5" by overseas developers, and Musk also publicly praised it.
Source: User-contributed newsletters
June 9, 2026
Khanmigo Product Observation: Capability Inventory of AI Learning Direction
Khanmigo is a GPT-4 AI learning tutor launched by Khan Academy. It takes Socratic teaching method as its core and guides students to actively think instead of passively accepting answers.
Source: Khanmigo official documentation
Re-understand Kagi: the core capabilities of AI search tools
Kagi is a paid search engine for users who value privacy and search quality. It has no ads and no tracking. It has built-in AI page summaries, FastGPT instant answers, Lenses custom search ranges and multi-LLM conversation assistants. It completely eliminates the interference of advertising on search results from the business model level.
Source: Kagi official documentation
Re-understand Julius AI: the core capabilities of AI data processing tools
Julius AI is a conversational AI data analysis platform. After users upload CSV/Excel files, they can complete data exploration, generate visual charts, run statistical models and execute Python code through natural language conversations.
Source: Julius AI official documentation
How to use Jasper: AI content platform for enterprise marketing teams
Jasper is an AI platform for enterprise marketing teams, providing AI Agents, brand voice, Jasper Studio and other capabilities. This article summarizes its capability stack, pricing and applicable scenarios.
Source: Jasper official documentation
Ideogram Capability Inventory: Selection Reference for AI Image and Design Teams
Ideogram is an image generator created by the former Google Imagen team. It is good at text rendering, poster design and branding materials.
Source: Ideogram official documentation
How to use InVideo AI: Enter prompt words to automatically generate a complete video with narration and subtitles
InVideo AI supports inputting text prompts to automatically generate complete videos. It has built-in AI scripting, AI dubbing, subtitles and 5000+ templates. It is a video creation tool for 25 million users. It starts for free. Plus $25/month.
Source: InVideo AI official documentation
How to use Hugging Face: the center of the AI open source ecosystem
Hugging Face is the central hub of the AI open source ecosystem, hosting models, data sets and Spaces demonstrations, and providing open source libraries such as Transformers and Diffusers as well as Inference Endpoints.
Source: Hugging Face official documentation
HeyGen Capability Inventory: Selection Reference for AI Video Team
HeyGen is an enterprise-level AI digital human video platform covering Avatar IV, Interactive Avatar, video translation and Video Agent. This article summarizes its capability stack, pricing and typical scenarios.
Source: HeyGen official documentation
How to use HubSpot: All-in-one CRM platform
HubSpot is an all-in-one CRM platform integrated with Breeze AI covering marketing and sales services. This article summarizes its functions, pricing and application scenarios.
Source: HubSpot official documentation
Helicone AI Product Observation: Capability Inventory in AI Data Processing Direction
Helicone AI provides LLM request monitoring, routing, and evaluation capabilities, and is suitable for production-level AI product teams that require cost control, stability management, and multi-model experiments.
Source: Helicone AI official documentation
What is Harvey AI: The AI legal platform for top law firms and corporate counsel
Harvey AI is an enterprise-level AI legal platform founded by the former OpenAI team. It is specially designed for top law firms and corporate legal affairs. It provides contract analysis, due diligence, regulatory research and litigation support. It has received strategic investment from OpenAI and Google.
Source: Harvey AI official documentation
Haystack Capability Inventory: Selection Reference for AI Agent Teams
Haystack is an open source LLM application framework produced by deepset. It supports RAG, document retrieval, question and answer system and conversational agent. Its modular pipeline design allows developers to quickly build production-level AI applications.
Source: Haystack official documentation
Understand Hailuo AI in one article: MiniMax’s video generation platform
Hailuo AI is MiniMax's AI video generation platform, covering Wensheng videos, Tusheng videos and creator workflow. This article summarizes its model iteration, capability highlights and pricing path.
Source: Hailuo AI official documentation
Grok function dismantling and applicable scenario analysis
Grok is an AI chat assistant launched by xAI, focusing on real-time X data access, Grok 4 reasoning and Imagine image/video generation. This article summarizes its capabilities, pricing and typical application scenarios.
Source: Grok official documentation
Groq capability analysis: How to use AI model training scenarios
Groq provides the industry's fastest AI inference speed through self-developed LPU hardware, up to 1000 tokens/second, and supports open source models such as Llama 4 and Qwen3.
Source: Groq official documentation
July 15, 2026
xAI open source Grok Build coding Agent harness: "Open source surprise" on the terminal programming track
On July 15, xAI announced the open source Grok Build coding agent and the harness engineering framework behind its terminal TUI, on which developers can build custom coding agent workflows. This is a key move to continue xAI's "open + cost-effective" strategy after Grok Code Fast 1, directly entering the terminal programming track led by Claude Code and Codex CLI.
Source: xAI (SpaceXAI) official news
July 28, 2026
Musk predicts Grok 4.6/4.7: 1.5T parameters will be launched in August, 2.1T parameters will follow closely
Musk announced the xAI next-generation model roadmap on July 28: Grok 4.6 is expected to be released around August 7, with a parameter size of 1.5 trillion, and significant improvements in SFT and RL training; Grok 4.7 with a parameter size of 2.1 trillion will be launched in a few weeks, hitting the upper limit of capabilities with a high-frequency burst.
Source: User-contributed newsletters