AI
The Labs Stopped Selling Models and Started Selling the Whole Workplace
The AI labs spent this week selling something new, and it is bigger than a model. Google shipped Gemini 4 Hydrogen alongside an orchestrator that mixes models inside a single session. Anthropic put self hosted Claude Managed Agents into private beta and watched its flagship top the independent scoreboard. OpenAI opened a security product line while publishing research on how hard its own models are to monitor. The common thread is plain. The product is now the workplace where the work happens, and the model is one piece inside it.
Google led with execution. Gemini 4 Hydrogen launched November 12 as an execution oriented member of the Gemini 4 family, priced at 1.50 dollars per million input tokens and 7.50 dollars per million output tokens across a 1 million token context window. It scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, ahead of Gemini 4 Argon at 50, and reaches 54 at its highest thinking setting. It finishes tasks 2.5 times faster than Gemini 3.8 Flash while more than tripling Terminal-Bench 4.1 scores, 57 percent against 18.2 percent. The telling number sits in the multi model story. Orchestrator in Google Antigravity switches between models like Carbon and Hydrogen within the same session, reaching near Carbon task success rates while consuming 24 percent less quota and completing tasks 19 percent faster on Terminal-Bench 4.0.
Anthropic split the frontier into planners and workers, and the scoreboard noticed. Claude Fable 5.2 launched November 9 and topped the Artificial Analysis Intelligence Index at 65, ahead of Gemini 4 Carbon at 60 and Anthropic's own Claude Opus 5.5 at 59. Pricing is 8 dollars per million input tokens and 40 dollars per million output tokens. Meta went the other direction the same day with Muse Ember at 0.10 dollars per million input tokens, built to execute repetitive agent tasks while a larger model plans. Anthropic named Jump Trading, Balyasny Asset Management and XTX Markets as quantitative research users, and Box as a user reviewing loan application files. The same week Anthropic opened self hosted Claude Managed Agents in private beta, letting enterprises run agent servers, credentials, sandboxes and session history in their own cloud or on premises. Claude Whiteboards and Claude Sheets arrived in beta, Claude Data went generally available, and Claude Meetings joined Google Meet calls in beta. The roadshow added a market angle, with press reports of a 70 to 78 dollar per share range for 825 million shares and a planned Nasdaq listing under the ticker ANT.
OpenAI turned its attention to defense and to watching its own creations. OpenAI announced Daybreak Red Team and Defense Factory on November 12. Daybreak Red Team probes web applications for weaknesses in early access, and Defense Factory automates finding, verifying and fixing vulnerabilities, with engineers approving patches. The same week brought GPT-6 Cyber, limited to approved customers in the Daybreak Red security program, which found 81 previously unknown high severity Firefox flaws in testing and built 948 working exploits against the Linux kernel. OpenAI also published research on November 10 examining why increasingly capable reasoning models become harder to monitor through their visible chain of thought. The lab is selling security and studying its own blind spots in the same breath, and both moves point the same direction.
The factory floor and the payment rail round out the week. Google DeepMind released Gemini Robotics 2.5 to robotics partners on November 10, and in a pilot on live Mercedes Benz production lines in Hungary and Germany, Apptronik Apollo 2 humanoid robots running the model sorted metal brackets into trays in vehicle order with a 96 percent success rate. Brave Search API now lets agents pay per query over the x402 protocol. Thinking Machines Lab released Inkling 2 Small on November 5, a 594 billion parameter model that uses only 17 billion per request, priced at 0.30 dollars per million input tokens, with weights promised under an Apache 2.0 license after outside safety testing. Baidu released ERNIE 5.5 Flash Base the same day with open weights.
The buying question has changed. It used to be which model is smartest. Now it is which model handles each task at the lowest price, and which workplace keeps the whole team in one room. Hydrogen for execution, Fable 5.2 for planning, Ember for grunt work, Orchestrator to route between them. Learn the routing, and the frontier gets cheaper every month. The scoreboard still matters, and the invoice matters more.
Quick answers
What is this story about?
The AI labs spent this week selling something new, and it is bigger than a model. Google shipped Gemini 4 Hydrogen alongside an orchestrator that mixes models inside a single session. Anthropic put self hosted Claude Managed Agents into private beta and watched its flagship top the independent scoreboard. OpenAI opened a security product line while publishing research on how hard its own models are to monitor. The common thread is plain. The product is now the workplace where the work happens, and the model is one piece inside it.
Why does this story matter?
The buying question has changed. It used to be which model is smartest. Now it is which model handles each task at the lowest price, and which workplace keeps the whole team in one room. Hydrogen for execution, Fable 5.2 for planning, Ember for grunt work, Orchestrator to route between them. Learn the routing, and the frontier gets cheaper every month. The scoreboard still matters, and the invoice matters more.
Sources
- Google Blog, Introducing Gemini 4 Hydrogen
- PYMNTS, Anthropic Claude Tops AI Rankings
- DDS Boston, AI News for Vibe Coders November 13
- Promact, Enterprise AI News This Week November 13
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.