AI/Tech Brief — June 25, 2026

TL;DR

• Google announced computer use capabilities for Gemini 3.5 Flash, enabling agentic UI navigation natively within the model.
• OpenAI revealed its first custom AI chip developed with Broadcom, a major step toward reducing its dependency on external hardware suppliers.
• Anthropic publicly accused Alibaba of illicitly extracting capabilities from its Claude models, highlighting growing tensions around AI intellectual property.

Key Stories

Google Introduces Computer Use in Gemini 3.5 Flash Google has updated Gemini 3.5 Flash to include computer use capabilities, allowing the model to autonomously navigate UI elements and perform tasks on a desktop. This brings built-in agentic features to their highly efficient Flash tier, positioning it strongly against Anthropic’s recent computer use releases. Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-computer-use-gemini-3-5-flash/

OpenAI Unveils First Custom Chip OpenAI has officially unveiled its first custom silicon, developed in partnership with Broadcom. This hardware milestone is critical for OpenAI’s long-term unit economics and marks a definitive move toward supply chain independence away from strict reliance on Nvidia. Source: https://techcrunch.com/2026/06/24/openai-unveils-its-first-custom-chip-built-by-broadcom/

Anthropic Accuses Alibaba of IP Theft Anthropic claims that Alibaba extracted capabilities from Claude AI models without authorization. This dispute underscores the industry’s struggle to enforce usage policies and prevent “model extraction” attacks by competitors building their own foundation models. Source: https://news.ycombinator.com/item?id=48664814

Cloudflare Launches Self-Managed OAuth for All Cloudflare has rolled out self-managed OAuth capabilities globally. This gives developers tighter control over user identity and access management directly at the edge, simplifying authentication workflows without relying on external third-party identity providers. Source: https://blog.cloudflare.com/oauth-for-all/

Claude Code Updates: Faster Streaming and /rewind Anthropic shipped updates (v2.1.191) for Claude Code, introducing a new /rewind command to resume conversations from before a /clear was issued. CPU usage during streaming has been reduced by ~37%, and several bugs involving background agent resurrection have been patched. Source: https://code.claude.com/docs/en/changelog

OpenAI API Adds Safety Usage Dashboard OpenAI’s latest platform update includes a new Safety Usage Dashboard that tracks blocked requests via safety_identifier metrics. Additionally, the chat-latest pointer has been updated to route requests to the newest ChatGPT Instant model snapshot. Source: https://developers.openai.com/api/docs/changelog

Quiet But Interesting

Superhuman Integrates Claude into Slack Superhuman’s latest update brings Claude directly into Slack channels. This integration allows teams to generate email drafts, summarize threads, and manage communications seamlessly within their chat workspaces without context switching. Source: https://www.superhuman.ai/

Large Language Models vs Small Language Models ByteByteGo published a deep-dive analysis comparing the architectural tradeoffs between massive LLMs and purpose-built Small Language Models (SLMs). The piece highlights when to prioritize efficiency over broad reasoning capabilities. Source: https://blog.bytebytego.com/

Half-Life 2 Ported to WebAssembly The iconic game Half-Life 2 is now fully playable natively within a web browser, showcasing the significant performance strides WebAssembly has made for rendering complex 3D environments without dedicated local binaries. Source: https://hl2.slqnt.dev/

Skip

• Claude Support, Sam Altman, Dario Amodei, Gemini CLI: No major updates or news from these channels today. Time better spent elsewhere.