Google Integrates Computer Use Capabilities into Gemini 3.5 Flash

Google has directly integrated its "Computer Use" capability into Gemini 3.5 Flash, enabling the model to operate computers, browsers, and mobile devices on its own. It scored 78.4 on the OSWorld benchmark, matching the performance of GPT-5.5. Through the Gemini API, developers can build agents for software testing and office automation.

This marks a major step forward in the practicality of "agents" — AI models that don't just generate text but actually operate real devices. The competition between Google and OpenAI over computer-use capabilities looks set to intensify.

OpenAI Publishes Research on How Agents Will Transform Work

In a new research paper, OpenAI has laid out how AI agents will transform the way we work. It shows that agents can take on longer, more complex tasks, unlocking productivity gains across job categories.

Agent technology is expanding AI's scope from one-off question answering to workflows that autonomously complete multi-step tasks. The paper is drawing attention as foundational data for enterprise adoption.

Meta Employees Warn Over the Rapid Rollout of AI Moderation

Employees at Meta are sounding the alarm that the rollout of AI-driven content moderation is moving too fast. As of 2025, Meta has replaced roughly half of its human moderation requests with large language models, and it aims to raise that figure to over 90% for certain content types by the end of the year.

Debate continues across society over the accuracy and fairness of AI moderation, and the fact that concerns are being raised from within the company is a signal worth watching.

AI Researchers Continue to Leave Google

Google is reportedly continuing to lose top AI researchers to competitors. The outflow of AI talent, centered on DeepMind, could erode the company's competitive edge.

The race for AI talent is intensifying, with researchers leaving labs one after another to join independent startups or rival companies.

Qualcomm Enters the Data Center Market — Dragonfly C1000

Qualcomm has announced a full-fledged entry into the data center market with a new processor called the "Dragonfly C1000." The company has built its strength in mobile, and its push into data center workloads including AI inference could reshape the competitive structure of the semiconductor industry.

Anthropic Claims Improper Access by Alibaba

Anthropic has claimed that Alibaba ran a "bold" campaign to gain access to the Claude AI models. The details, reported by the WSJ, highlight fresh concerns around the protection and security of AI model intellectual property.

Safeguarding models and preventing unauthorized access is becoming a challenge for the AI industry as a whole.

(Reference) Sampler Technique Dramatically Boosts Small-Model Coding Performance

According to research making the rounds in Reddit's LocalLLaMA community, a new combination of samplers and verifiers can lift the coding performance of small 0.5B-parameter models to a level rivaling 2–4B class models. That said, there are practical hurdles: the verifier model requires additional training, and VRAM requirements double.

---