Anthropic previews Model Hardware Standard for agent-operated equipment

#1 𝕏

Anthropic announced the first phase of the research preview for Model Hardware Standard (MHS), described as a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing.

Also covered by: @Anthropic

#2 𝕏

Google AI announced Gemini Omni 1.1 Flash, a new multimodal model for video generation and editing with Veo creative controls, 4K upscaling, first/last frame control, and fast 360p drafting. It can extend scenes using 10 seconds of context from the original video, compared with Veo’s 1 second.

Also covered by: @Google AI, @Philipp Schmid, @Logan Kilpatrick, @Philipp Schmid

#3 𝕏

Qwen3.8-Flash is now live on OpenRouter, supporting coding assistants, agentic workflows, and long-video understanding through one API call.

Also covered by: @Qwen

#4 𝕏

clem 🤗 says Hugging Face announced Microduck, a tiny $399 open-source robot that can be taught new tricks with reinforcement learning. It can walk, pick things up, get back up after falling, and roller-skate.

#5 in

Dharmesh Shah announced YouSpot, an AI-native CRM from HubSpot Next for one-person businesses that builds business context, surfaces follow-ups and opportunities, and supports MCP-connected AI apps and cloud agents. The early product uses pay-as-you-go pricing, with a $1/month introductory price for the first 1,000 customers; 423 slots were already taken.

Also covered by: @Dharmesh Shah

#6 𝕏

Users can now create new web apps with Cursor, store the code with Origin, and deploy them to Vercel.

#7 𝕏

Garry Tan said he cut GStack’s token load by 50% with no reduction in capability. The repository is available at github.com/garrytan/gstack.

#8 𝕏

Thinking Machines shared a guest post by UIUC and Bridgewater researchers, written in collaboration with its team, arguing that upfront expertise and effort in cleaning data and aligning the RLVR reward function resulted in a model described as state-of-the-art on a complex task.

#9 𝕏

Guillermo Rauch announced that vgpu.sh is available, describing it as a fully agent-native development tool designed for agents, not humans alone. He also noted that @matinotfound and team had been shipping WebGPU creative productions.

#10 𝕏

LlamaIndex 🦙 released native spreadsheet extraction for LlamaParse, which reads raw cells and maps data to a user-provided schema instead of flattening workbooks into text. The feature is available in beta on the agentic_plus tier and supports .xlsx, .xls, and .csv files.

#11 𝕏

Ali Ghodsi recommended branching databases on Neon Lakebase to prevent agents from permanently wiping data, sharing the command `neonctl branches create --name newbranch`.

#12 𝕏

Marily Nika says AI has made disposable, happy-path prototypes nearly free, but demos are not products. A viable MVP must handle real edge cases and include evaluations, monitoring, and trust.

#13 𝕏

Google DeepMind announced a pilot of double-blind evaluations for frontier AI, using a secure environment that reveals neither test prompts nor model weights. The company says the approach—characterized as an industry first—is intended to keep external safety and performance evaluations private, robust, and trustworthy.

#14 𝕏

Josh Woodward announced a program that lets users buy a book, add it to Notebook, and apply the author’s lessons to a project. The program was co-created to help authors and publishers reach new and more engaged readers.

#15 📝 Anthropic News

Expanding our support for scientists - Anthropic is opening 10,000 seats for scientists to join a new Claude team plan (standard seats free for one year; premium seats with 5x usage limits available for $15/month) that PIs at academic or nonprofit research institutions can register for and extend to their labs. They’re also expanding the AI for Science program to fund researchers beyond biology—offering up to $50,000 in credits per project—while keeping biology and chemistry on Opus-class models, continuing to block professional bio/drug queries on Claude Fable, and partnering with the U.S. government to pilot controlled Mythos-class access (first participants enrolled).

Also covered by: @Claude

#16 𝕏

Google Research announced the planetary prediction engine, an experimental research capability that automates global geospatial modeling workflows. It is designed to simplify planetary-scale analytics for applications ranging from predicting public health outcomes to improving food security.

#17 𝕏

Guillermo Rauch described CMA as an elegant architecture that manages the “hard parts” of agents while preserving customization, including fully custom UIs with Chat SDK and custom tools using your own functions on Vercel.

#18 𝕏

Bolt released prompt queueing, allowing users to send another message while Bolt is still building. Prompts queue and send automatically.

Also covered by: @bolt.new

#19 𝕏

Agentic trading was demonstrated on Perplexity Computer using the @public connector.

#20 𝕏

Marily Nika shared a command for any AI tool to create five markdown files capturing a user’s business, writing voice, recurring workflows, current priorities, and other personal context, while asking questions to fill gaps. She characterized the process as taking “Thirty minutes” to make AI tools interchangeable by providing portable context for any model.