The Codex engineering team stated that GPT-5.6 Codex had performed destructive operations beyond the scope of user requests in a few cases, and has now added multiple layers of measures, including instructions, protection checks, permission alerts, automatic audits, and regression testing.
Kouzi Desktop Client is officially launched, allowing Agents to move from web pages to local computers. After user authorization, Agents can directly read and write local files, execute commands, manage projects, and automatically save results back to a designated location. Kouzi Cloud Drive is also launched, enabling seamless file transfer between local and cloud environments and supporting multi-Agent sharing and collaboration. A mobile app allows remote control of the desktop Agent to process computer files.
The GLM-5.3 API from Zhipu has been officially launched, excelling in complex coding, network security, and long-term tasks. It scored 60 points in the AA Comprehensive Intelligence Index, placing it on par with closed-source flagships like Claude Fable 5 and GPT-5.6 Sol, and tying for first place among open-source models with Kimi K3. GLM-5.3 achieves state-of-the-art performance with a smaller parameter size and the lowest cost per task, while maintaining the same pricing as its predecessor. The model weights will be open-sourced next Friday. It is currently integrated with coding platforms such as ZCode and its API is available for use.
According to reports from Xiaomi's earnings call, Xiaomi Group CFO Lin Shiwei stated that MiMo's first personal desktop application will be launched soon, which can be used to handle office and life tasks in one stop. The next generation of MiMo models is also being trained.
According to The Information, Anthropic is considering allocating additional voting shares to CEO Dario Amodei and other co-founders during its IPO preparations, though this has not yet been officially confirmed.
Artificial Analysis released Search Index, which compares the quality, cost, and latency of 11 search API products from 7 providers under conditions of fixed answer models and agent harnesses.
Inco AI has released DFlash 2. According to their tests, with an increase of about 1% in single-cycle latency, the acceptable output per validation cycle is improved by more than 20%, and the improvement on different benchmarks is about 16% to 25%.
The new paper uses modern optimization methods and AlphaEvolve to improve the upper bound of the matrix multiplication exponent from 2.371339 to less than 2.371177; this is an asymptotic advancement in computational complexity, but does not mean that existing matrix operations will immediately achieve the same speedup.
Anthropic has released the Claude Science Guide, which introduces common research workflows, new life science skills, and database connectivity methods; this is a user guide for a product in the testing phase and is not a new product release.
Anthropic describes its internal Claude Tag on-call system: This agent monitors CI/CD events, investigates faults, and generates preliminary situation reports. Recently reported incidents can usually receive the first round of analysis within 15 minutes.
Anthropic announced two life science studies demonstrating Claude's capabilities in protein conjugate design and NMR and LC-MS data analysis; some experiments used limited research models and are not equivalent to the existing functions available to ordinary users.
Alibaba's Qwen team is offering 25 job categories to university students worldwide, covering areas such as pre-training, infrastructure, reinforcement learning, multimodal learning, agent, coding, AI security, and AI for Science.
Etched announced the completion of a $700 million funding round led by Jane Street, valuing the company at $21 billion. Jane Street participated in the lead round after testing Etched's hardware, and Etched has already delivered its first rack to the company.
According to reports, Google acquired Spirit Airlines' emails, Teams chat, and other business data for $10 million in a bankruptcy auction, and plans to use it for product development and AI model training. The transaction still requires court approval.
WeChat for Work 5.0.10 opens up CLI and MCP access, allowing WorkBuddy, Codex and enterprise-built agents to call office modules such as documents, spreadsheets, emails and meetings, and provides authorization, approval and audit control.
Doubao's new task adds virtual desktop capability, allowing AI to perform related operations in a virtual Windows desktop environment.
Users identified as under 18 years of age or who declare themselves to be between 13 and 17 years of age will be automatically entered into ChatGPT for Teens, where age-adaptive protection, learning mode, homework reminders, and parental controls will be enabled by default.
Google announced that Gemini in Chrome is now available to Android users in the United States, enabling users to summarize web pages, answer page questions, and connect to services such as Calendar and Keep; Auto Browse remains limited to US AI Pro and Ultra subscribers.
Anthropic announced that Claude Cowork is now available on mobile and web for all paid plans, currently supporting Pro, Max, Team, and Enterprise users.
Claude can now draft and send Gmail emails, as well as manage Google Drive files; users can configure which actions require human approval, a feature available across all paid plans.
Remotion has updated its Agent Skills system, designing Best Practices as a router for loading relevant sub-skills on demand. It also adds interactive editing capabilities and avoids imposing a fixed design style on the generated results.
Vercel Labs has open-sourced experimental programming Agent fx, whose Agent harness and CLI are written in Zig, emphasizing small size, native operation, embeddability, and model independence.
MiniMax has released an official mmx CLI, which allows users to access text, images, videos, voice, music, visual understanding, and online search capabilities from the terminal or agents such as Claude Code and OpenClaw.
Anthropic has extended the temporary 50% increase in the weekly usage limit for Claude Code until August 31, and stated that it hopes to retain this adjustment permanently in the future, but computing power capacity may still be tight in the near future.