2026-11-17
By Vadym · Generated with AI, curated by me
Google finished rolling out its three-tier Gemini 4 family this week, a rival lab shipped open weights exactly on the date it promised, and two separate stories made the same point from different angles: access to a model, a platform, or a product lead can be revoked faster than it was granted.
Google DeepMind released Gemini 4 Hydrogen, a version adapted from its frontier Gemini 4 Carbon model and built for everyday coding and agentic work rather than deep reasoning. It handles up to 1M tokens of combined context and leads on benchmarks like Google's SWE-sweep (13.5% vs. under 6% for comparison models), though it trails Claude Sonnet 5.5 and GPT-6.1 Terra on terminal and OS-navigation tasks. It's rolling out across the Gemini app, API, Antigravity, and Search.
DeepMind released Gemini Robotics 2.5, its vision-language-action model, trained and tested across more than 200 manipulation tasks. In a Mercedes-Benz proof of concept, Apptronik's Apollo 2 humanoid running the model hit a 96% success rate on automotive kitting. DeepMind also reports inference-time safety reasoning pushed jailbreak resistance to 99.5%, up from 82% on its prior baseline. It's rolling out to robotics partners only, not generally available.
LangChain raised $300 million in a Series C led by Alphabet's CapitalG, with GIC, DFJ Growth, IVP, and Sapphire Ventures joining, putting the agent-tooling company at a $3 billion valuation — 2.4x its $1.25 billion mark from October 2025. LangChain says more than 7,000 organizations now use LangSmith, its agent-monitoring platform.
Cat Wu, who led product for Claude Code and Cowork at Anthropic, announced on November 5 that she's moving to OpenAI to run product management for its Core Products team. Her last day at Anthropic was October 30. She'd joined Anthropic in August 2024 and became one of the most visible faces of Claude Code's product direction.
Three of today's six stories are about who controls access: to a model (OpenAI cutting off Cursor), to the agent-tooling layer underneath models (LangChain's raise), and to the people who know how to build with them (Cat Wu's move). Even Google's tiered Gemini 4 rollout is an access story — pick the tier you can afford. The race stopped being about whose model is smartest a while ago. It's about who decides who gets to use it.
— Boba
Curated by Vadym