The New Stack

Apple is splitting up its AI stack. Instead of rolling out the same system worldwide, the company reportedly built a China-only model with help from Alibaba to navigate the country’s tough regulatory hurdles. Reuters reported Friday that Apple’s new model is expected to work alongside Alibaba’s Qwen and possibly technology from Baidu. Neither Apple nor its Chinese partners has explained which model will handle each part of Apple Intelligence.

Alibaba's new model promises Opus 4.6-level performance on your laptop Alibaba recently made the open weights of its 2.4 trillion parameter Qwen3.8 model available. That’s a massive model, and its benchmarks put it in direct competition with closed frontier models from American labs. But what’s maybe even more interesting is that on Friday, Alibaba also made a dense 27 billion parameter version of Qwen 3.8 available under the Apache 2.0 license.

Z.ai released GLM-5.3 on Friday, a coding and agent model built from the same base model as GLM-5.2. Developers can already use GLM-5.3 through Z.ai’s GLM Coding Plan with Claude Code, Cline, OpenCode and Codex. Direct API access is still listed as “coming soon,” because Z.ai plans to release the model weights after two weeks of hardening and safety testing.

Most organizations that know they should sign their images still don’t. Not because they disagree, but because the path to doing it well has been too long. The result is a delivery pipeline built on trust that nobody can verify. Unsigned container images create an open door for attackers at every stage of the delivery pipeline. Malicious images masquerade as legitimate packages, waiting to be pulled by an unsuspecting team.

Google has a new Gemini model, and no, it is not Gemini 3.5 Pro. Gemini 3.7 Flash launched Thursday as the company’s new workhorse for coding agents and automated business workflows, just three weeks after Gemini 3.6 Flash became available. The quick turnaround may cause some version fatigue, but Google says the new model is better at writing code and handling the longer workflows that keep agents on track.

ChatGPT can now remember what you did on your Mac — without screenshots Dima Solomin via Unsplash. OpenAI is launching a new feature for ChatGPT Work and Codex on macOS that sounds quite useful but may also make you feel a bit uneasy. Computer History is a new optional feature for Pro, Business, and Enterprise users that, with your permission, can watch your day-to-day activity on your computer across apps and websites and turn that into memories and a timeline.

After it joined Project Glasswing and gained access to experiment with Mythos Preview, security and AI company Rubrik learned it didn’t have the engineering capacity to keep up with the model’s vulnerability-hunting capabilities.

DeepSeek on Thursday open sourced the DeepSeek Harness, a new agent runtime for developers. The Node.js-based harness is now available on GitHub as a developer preview and under an MIT license. Clearly, there is some interest in the new harness. Within only a few hours, the repo picked up more than 33,000 GitHub stars, and that number is climbing quickly. Theirs is already a thriving ecosystem of community plugins.

Every token has a price. The problem is that most AI systems don’t reveal the bill until they reach production. By then, what looked like an intelligent application has become an expensive one—not because the model is flawed, but because the architecture is inherently inefficient. The most expensive bug in your AI application might not be a hallucination; it might be the thousands of unnecessary tokens your users never notice.

Code review is becoming the most important decision-making surface in software engineering, and it’s outgrowing the diff. The assumption was always that code review exists to catch bugs. Its real job is taste, judgment, and applying organizational processes: Is this the right thing for our product?