The Day in AI
In short
- Google released Gemini 4 Hydrogen, a cheaper Gemini 4 model for coding and agents, and moved all Search users onto it the same day.
- Google reports that Hydrogen scores 57% on Terminal-Bench 4.1. Gemini 3.8 Flash, the model it replaces, scored 18.2%.
- Anthropic now lets customers run the whole Claude Managed Agents service on their own infrastructure, with agents and session history stored there.
- Jensen Huang has resigned as Nvidia's CEO, president and board member, according to reports, and CFO Colette Kress is interim CEO.
- Google and Samsung opened preorders for Gemini smart glasses without a display, starting at $395 and reportedly shipping in December.
The day in brief
Google released Gemini 4 Hydrogen on 12 November at $1.50 per million input tokens and $7.50 per million output tokens, and made it the model behind Search for all users.
Hydrogen replaces Gemini 3.8 Flash and sits below the Carbon and Argon frontier models in the Gemini 4 family. It has a 1 million token context window with flat pricing, so long prompts cost the same per token as short ones. It's available in AI Studio, the Gemini API and Antigravity. Google rolled it out to Search users globally on the same day and calls it the biggest model upgrade to Search yet. In Search, Hydrogen improves query fan-out, where Search splits a question into several subqueries that run at the same time, and Google says this surfaces a wider range of links.
Google reports 57% on Terminal-Bench 4.1 and 58.8% on HLE-Diamond, and says Hydrogen finishes tasks 2.5 times faster than 3.8 Flash at its default high setting. Vals ranks it first overall on the Vals Tax Agent Bench at 52.3%. On the Vals Legal Research Bench it ties for first at 55.8%. On arena.ai's Agent Arena it sits fifth.
Google also added Orchestrator to Antigravity, which is rolling out to AI Ultra subscribers. Orchestrator pairs Carbon and Hydrogen in one session and switches between them in both directions. Carbon can explore the code read-only and write a plan, then Hydrogen makes the edits and runs the tests, and it hands back to Carbon if it gets stuck. Google reports near-Carbon task success on Terminal-Bench 4.1 with 24% less quota, and it finishes 19% faster.
Anthropic launched self-hosted Claude Managed Agents, available through customers' Anthropic account teams. Until now Managed Agents ran as a hosted service, and self-hosted sandboxes covered only the environment where code executes. Customers can now run the whole service and keep agents, sessions and session history on their own infrastructure. The server sends model requests to the provider the customer chooses, which is Anthropic by default.
Also in the news
- Jensen Huang has resigned from Nvidia after reports of affairs, including one with a vice president who reported to him, according to reports. Harvey C. Jones became the company's first board chair while the board looks for a permanent CEO.
- Google and Samsung opened preorders for Gemini glasses in Warby Parker frames from $395 and Gentle Monster frames from $449, and Ars Technica reports that faces are redacted before video is processed.
- Baidu released ERNIE Lightning, a reasoning model for Baidu Search that uses Baidu Sparse Attention to compress the KV cache, and Baidu says it cuts inference engine cost by more than 30%.
- ByteDance released Seedance 2.5 Flash on Volcano Engine, which it says is twice as fast as Seedance 2.5 and costs about 0.99 yuan per second of 720p video.
- LangChain has more than doubled its valuation to $3 billion, according to The Wall Street Journal.
- Agile Robots, a German robotics startup, is reportedly valued at $8.5 billion after a new funding round.
- Oura is reportedly acquiring the medical AI startup Counsel Health for $625 million.
- Meta's infrastructure chief Santosh Janardhan is leaving the company, according to The Information.
- The US Commerce Department is finalizing its long-awaited AI remote access rule, according to The Information.