In partnership with
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
Today:
Kimi Launches HighSpeed Coding Mode
Z.ai Releases Open-Source GLM-5.2
SpaceX Buys Cursor for $60 Billion
BytePlus Debuts Budget AI Video Tool
Factory 2.0 Automates Software Engineering
Kimi.ai has introduced "HighSpeed," a turbocharged execution mode for its recently launched open-source multimodal coding model, Kimi K2.7 Code. Designed to drastically reduce latency in agentic workflows and interactive programming, this new mode accelerates generation by up to six times, delivering near-instantaneous code completion without sacrificing accuracy.
Important Details:
Unprecedented Speeds: The HighSpeed mode outputs around 180 tokens per second (tok/s) on standard-length coding tasks. For tasks with shorter contexts, generation speeds ramp up to a massive 260 tok/s.
Current Availability: The feature is actively rolling out to Kimi Code Beta API developers and Kimi Business users. Because of current hardware capacity constraints, access remains somewhat limited for now.
How to Get Access: No special invitation is required to try it. Developers can join the open Beta Program directly at kimi.com/code/beta to get in line for access.
The Broader Mission: Kimi.ai is using this release to double down on its open-source mission, noting that "open intelligence should be instant, affordable, and borderless." They plan to expand access steadily as more compute capacity becomes available.
Chinese AI startup Z.ai (formerly Zhipu AI) has officially released GLM-5.2, an immense 753-billion parameter large language model engineered for autonomous software engineering. Released fully under an unrestricted MIT open-source license, GLM-5.2 provides a powerful way for enterprises to host frontier-level AI on local infrastructure, bypassing geopolitical export restrictions and costly vendor lock-in.
Important Details:
Massive Context: GLM-5.2 boasts an incredibly stable 1-million-token context window.
Performance Benchmarks: The model scored 62.1 on SWE-bench Pro and 74.4% on FrontierSWE, decisively edging out rival proprietary models like GPT-5.5.
Hardware Innovation: It introduces an architectural breakthrough called "IndexShare," which reuses the identical indexer across sparse attention layers. At the maximum context length, this reduces compute FLOPs by 2.9 times.
Reasoning Modes: It features newly developed "High" and "Max" thinking modes, allowing users to trade processing speed for deeper, complex reasoning on agentic coding tasks.
Pricing: For developers who prefer not to self-host, Z.ai’s API subscription tiers start at just $12.60 a month (Lite plan), presenting a massive cost saving (roughly 1/6th the cost of comparable western models).
Following its blockbuster $2 trillion IPO and its February merger with Elon Musk’s xAI, SpaceX is acquiring Anysphere—the startup behind the wildly popular AI programming assistant Cursor. The $60 billion all-stock deal is a highly strategic move designed to immediately catapult xAI into the lucrative enterprise AI coding market and catch up to industry leaders Anthropic and OpenAI.
Important Details:
The Deal: Valued at $60 billion, the all-stock transaction is expected to close in Q3 of 2026. SpaceX utilized its massive post-IPO valuation to quickly secure the buyout rather than paying an alternative $10 billion partnership fee.
Mutual Benefits: Anysphere (Cursor) has previously cited a lack of compute power as a bottleneck to its growth; joining SpaceX grants it immediate access to a massive stockpile of AI chips. In return, xAI absorbs desperately needed top-tier engineering talent and Cursor's sophisticated joint-model training infrastructure.
Market Dominance: Cursor is a juggernaut in the B2B space, currently generating roughly $2.6 billion in annualized revenue. The acquisition firmly positions SpaceX to capitalize on AI-assisted coding, which is currently one of the few reliably profitable sectors of the generative AI boom.
🧠RESEARCH
OmniDirector lets video generators copy camera movement from reference footage, including cuts between multiple shots. It turns camera paths into moving grid videos, separating motion from people and scenery. Trained on one million examples, it beat competing systems in accuracy and reduced copying, though longer videos remain difficult to manage.
The survey argues that AI is moving from chatbots that answer questions toward digital colleagues that remember work, use tools, and finish tasks over time. Workspaces keep files and progress, while reusable skills store procedures. Reliable systems still need safety limits, memory, checking, recovery, and tests based on finished work.
JoyAI-VL-Interaction is an open AI model. It continuously watches video and decides each second whether to speak, stay silent, or request help. The system handles hours of live video with delays below one second. Testers preferred it over Doubao and Gemini, but the study was small and mistakes still happen.
📲SOCIAL MEDIA
🗞️MORE NEWS
BytePlus Seedance 2.0 Mini ByteDance has released Seedance 2.0 Mini, a tool that quickly creates videos using pictures, short clips, or sound recordings as a starting point. It is built for daily video creation at half the cost of their original system.
Factory 2.0 Factory 2.0 is shifting from basic digital assistants to an automated "software factory" that manages the complete process of building computer programs. The platform allows companies to run multiple artificial intelligence systems safely within their own private computer networks, learning from its own mistakes to improve. Instead of typing out code, human engineers will increasingly focus on directing these self-improving systems.
Anthropic Claude Code Expertise A study of Anthropic's Claude coding assistant shows that people with deep knowledge of a specific topic get much better results than beginners. While humans decide what needs to be built, the computer handles the actual building process, meaning you do not need to be a professional programmer to create working software. Users are spending less time fixing broken code and more time completing highly valuable projects.
Microsoft Copilot Cowork Microsoft is moving its Copilot business tool to a pay-for-what-you-use pricing setup to manage growing demand. To keep prices down, the company might offer DeepSeek, an open-source system whose internal code is freely shared with the public, as a cheaper alternative.
ByteDance AI Chips Procurement ByteDance is negotiating to buy at least 50,000 computer chips from Chinese manufacturer Iluvatar CoreX to run its automated chat programs. These chips will handle inference, which is the less demanding process of answering user questions after a system has already been built.
