[Dev Catch Up # 116] - OpenAI & HuggingFace Security Incident, Gemini 3.6 Flash, Qwen3.8, Grok For Excel, Cursor Router, Andrew NG's OpenWorker, CoWork's feature - Record a Skill, OpenShip and more!
Bringing devs up to speed on the latest dev news from the trends including, a bunch of exciting developments and articles
Welcome to the 116th edition of DevShorts, Dev Catch Up.
For those who joined recently or are reading Dev Catch Up for the first time, I write about developer stories and open source, partly based on my work and experience interacting with people all over the globe.
Thanks for reading Dev Shorts! Subscribe for free to receive new posts and support my work.
Some recent issues from Dev Catch up:
Join 8900+ developers to hear stories from Open source and technology.
Must Read
Last week, Hugging Face disclosed a security incident. An AI agent had compromised its infrastructure. OpenAI later found that its models were behind the incident. It happened during an internal cybersecurity test. The models found a way to access the internet from the isolated environment. They then accessed Hugging Face’s systems to get benchmark answers. Check OpenAI’s post for more details.
Google has released three new Gemini models. Gemini 3.6 Flash is built for coding, knowledge work, and multimodal tasks. It uses fewer output tokens than previous model. Gemini 3.5 Flash-Lite is built for fast, high-volume tasks. Gemini 3.5 Flash Cyber works with CodeMender to find and fix security vulnerabilities. Check Google’s post for more details.
Qwen has announced Qwen3.8, an upcoming 2.4T-parameter model. Qwen says the model will be released with open weights soon. Qwen3.8-Max-Preview is already available through selected Qwen platforms. Check Qwen’s post for more details.
Anthropic has added an interesting feature to Claude Cowork. There is a feature named Record a Skill. You can now record your screen while completing a task and explain the steps. Claude turns the recording into a reusable skill. It can use the skill to repeat the task later. Check Claude’s post for more details.
OSS Highlight of the Week
This week, we are featuring Openship. It is an open source, self-hosted platform for deploying apps. You point it to a repository, and it detects the stack, builds the app, and deploys it. You do not need to prepare configuration files, pipelines, or YAML files. It also manages databases, domains, SSL, and backups. Check the GitHub repo for setup details.
Good to Know
Sakana AI has released Fugu-Ultra v1.1 and Fugu-Cyber. Fugu-Ultra v1.1 is built for coding, agentic tasks, and advanced reasoning. Fugu-Cyber is a multi-agent system that can analyze vulnerabilities and create detection rules. Access to Fugu-Cyber requires approval. Check Sakana AI’s posts for more details.
NVIDIA has released Cosmos 3 Edge, a 4B-parameter model for robots and vision AI agents. It can understand scenes and generate robot actions on edge devices. Developers can adapt it for their own robot policies. Check NVIDIA’s post for more details.
Cisco has released Antares-350M and Antares-1B. These open-weight models can find files that may contain vulnerabilities. They run locally, so sensitive code stays within the environment. Check Cisco’s post for more details.
Poolside has released Laguna S 2.1, an open-weight coding model. It has 118B total parameters, 8B active parameters, and a 1M-token context window. It is built for long coding and agentic tasks. Check Poolside’s post for more details.
Notable FYIs
XAI has released Grok for Excel, an AI assistant that works inside Excel. You can now complete Excel tasks using simple prompts. It can write formulas, answer questions about selected data, create charts, and rerun scenarios. Check SpaceXAI’s post for more details.
Andrew Ng has released OpenWorker, an open source AI coworker that runs on your desktop. You can give it a task, and it completes the work for you. You can connect it to models from different providers or run local models with Ollama. Check the GitHub repo for setup details.
Cursor has launched Cursor Router. It routes each coding request to the right model. Simple tasks go to lower-cost models, while harder tasks go to stronger reasoning models. This helps teams reduce token costs. Check Cursor’s post for more details.
Unsloth now supports AMD hardware. Developers can train and run more than 500 AI models on AMD GPUs. It works on Windows, WSL, and Linux. Some Qwen and Gemma models can be trained with 3 GB of VRAM. Check Unsloth’s post for more details.
Qwen has released Qwen-Image 3.0, its latest image generation model. It can create detailed layouts such as newspapers, storyboards, and UI designs. It can generate small readable text and supports 12 languages. Check Qwen’s post for more details.
That’s it from us with this edition. We hope you are going away with a ton of new information. Lastly, share this newsletter with your colleagues and pals if you find it valuable. A subscription to the newsletter will be awesome if you are reading it for the first time.