majlis@daily — 11 Nov 2026 — Daily AI News — AI Majlis Daily
majlis@daily:~$ cd ~ / articles / 2026-11-11
AI Majlis Daily

Daily AI News — 11 Nov 2026

# Wednesday, 11 November 2026 · The AI stories that matter worldwide, ranked
# jump to a story
01Google ships Gemini 4 Carbon, its strongest frontier model [Industry]Google's deepest-reasoning model goes live for developers, enterprises and Ultra users.02Anthropic brings Claude into the office with Meetings and work tools [Industry]Claude moves from coding sidekick to an active teammate for knowledge workers.03Gemini Robotics 2.5 hits 96% on a Mercedes-Benz assembly line [Robotics]Google DeepMind's new vision-language-action model moves robots out of the lab.04DeepMind's AI formally verifies Signal's cryptographic protocol [Research]AI agents mathematically prove the code protecting billions of private chats.05Meta's Muse Ember delivers Haiku-level smarts at a third of the cost [Models]A 10-cent model built for the repetitive grunt work AI agents do all day.06Mistral open-weights Large 4, a trillion-parameter European model [Open Source]Le Chonk arrives fully open, built for coders and defensive security.07Mercor raises $750M at a $20B valuation to evaluate AI agents [Funding]The AI training-data company doubles its value as agent evaluation goes enterprise.08Rackspace and Together AI sign $380M GB300 inference deal [Infrastructure]A multiyear bet on dedicated NVIDIA clusters for managed AI inference.09Path Robotics raises $300M to scale AI welding robots [Robotics]Physical AI for manufacturing lands a Series E at a $2.45B valuation.10OpenAI launches Businesses in ChatGPT for small teams [Industry]A dedicated home for companies to run ChatGPT together.
#01 · Industry

Google ships Gemini 4 Carbon, its strongest frontier model

# Google's deepest-reasoning model goes live for developers, enterprises and Ultra users.
> the_news

Google made Gemini 4 Carbon generally available on November 10 to developers, enterprises, scientists and consumers. Positioned above Gemini 4 Argon, Carbon is tuned for long-horizon agentic work: complex multi-step coding and debugging, open-ended scientific research and frontier cybersecurity. Google says it burns more tokens to explore, reason and run extended verification loops, and ranks #1 on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at over 2x lower cost per task than the next best external model. It is in the Gemini API, Google AI Studio, Antigravity, the Gemini app and AI Mode in Search.

> why_it_matters

Carbon pushes the frontier on the tasks that matter to builders: sustained agentic coding, scientific discovery and defensive security. Pairing top-tier accuracy with lower cost per task lets teams run heavier workloads without blowing budgets, and tightens the race with OpenAI and Anthropic.

> real_life_use_case

A platform engineering team points Carbon at a sprawling migration, letting it debug across dozens of files and run its own verification loops overnight, then reviews a clean, tested pull request in the morning instead of hand-fixing each module.

#02 · Industry

Anthropic brings Claude into the office with Meetings and work tools

# Claude moves from coding sidekick to an active teammate for knowledge workers.
> the_news

On November 10 Anthropic launched a suite of AI office-productivity tools aimed at knowledge workers, part of its push to broaden paying customers ahead of a widely anticipated IPO. The headline feature, Claude Meetings, lets professionals add Claude to video calls as a visible participant: it listens, starts work while people are still talking, lays out a proposed plan and timeline, answers questions in the call chat, and pulls data from connected work apps, with results surfacing in claude.ai afterward. Claude Meetings is in beta for Team and Enterprise plans, starting with Google Meet.

> why_it_matters

Anthropic wants to do for every corporate role what Claude Code did for programmers. Turning Claude into a team member that acts during meetings, not just after, could reshape how non-technical teams work and expand the market well beyond engineering seats.

> real_life_use_case

During a quarterly planning call, a product team adds Claude as a participant; it quietly pulls last year's numbers, drafts the project timeline as colleagues debate, and has a first-pass plan waiting in claude.ai before the meeting ends.

> source

Fortune ↗

#03 · Robotics

Gemini Robotics 2.5 hits 96% on a Mercedes-Benz assembly line

# Google DeepMind's new vision-language-action model moves robots out of the lab.
> the_news

Google DeepMind introduced Gemini Robotics 2.5 on November 10, its most capable vision-language-action model yet. It widens both the range of skills a robot can learn and how fast it learns them, tested across more than 200 distinct tasks from using everyday tools to handling delicate, flexible objects. In a real-world proof of concept on Mercedes-Benz production lines, the model helped Apptronik's Apollo 2 humanoid pick and pack automotive parts with a 96% success rate. DeepMind reports 30% to 100% higher success rates than prior baselines and tasks completed 20% to 30% faster, plus new safety benchmarks against robotics jailbreaks.

> why_it_matters

Reliable dexterity on a live car assembly line is a real step toward robots doing monotonous physical work. Higher success rates and faster task completion are what move humanoids from demos to deployable factory-floor coworkers.

> real_life_use_case

On a production line, a humanoid running Gemini Robotics 2.5 handles the repetitive pick-and-pack of parts with near-perfect accuracy, freeing human technicians to focus on the intricate build steps that still need a person's judgment.

#04 · Research

DeepMind's AI formally verifies Signal's cryptographic protocol

# AI agents mathematically prove the code protecting billions of private chats.
> the_news

Google DeepMind revealed AlphaProof Code on November 10, an agentic framework that formally verified key security properties of the actual code behind the Signal protocol, used in Signal and Google Messages. Human experts translated Signal's security goals into formal specifications, then agents powered by Gemini Carbon generated over six million lines of Lean to prove the code correct. The system flagged just 1% of the codebase for human review and let Lean check the rest automatically. The whole verification took four weeks, a task that historically takes expert teams years, and it is the first time a widely deployed cryptographic protocol has been formally verified at this depth.

> why_it_matters

Formal verification is the gold standard of security assurance but is painfully slow and rare. Doing it on real, shipping crypto code in weeks instead of years points to a future where critical software can be mathematically proven safe at practical speed.

> real_life_use_case

A security team maintaining a messaging app runs AlphaProof Code against its encryption layer, getting machine-checked proofs that the protocol holds, and a short list of the few spots that genuinely need expert eyes rather than a months-long manual audit.

#05 · Models

Meta's Muse Ember delivers Haiku-level smarts at a third of the cost

# A 10-cent model built for the repetitive grunt work AI agents do all day.
> the_news

Meta Superintelligence Labs launched Muse Ember on November 9, its smallest and most efficient model, designed for the steps agents repeat constantly: tool calls, coding subtasks, routing, extraction and long-document parsing. It is natively multimodal across text, images, audio and video, with a 1M-token context window and up to 512K output tokens. On Artificial Analysis's Intelligence Index it scores 36, matching Claude Haiku 5.5 at roughly a third of the cost per task and beating GPT-6 Luna. Priced at $0.10 per million input and $0.50 per million output tokens, it is available today on the Meta Model API and in Muse Code.

> why_it_matters

Cheap, capable execution models paired with bigger planners are becoming the standard way to run agents affordably. At about $0.14 per task, Ember makes high-volume agent workloads economical enough to embed in nearly any product.

> real_life_use_case

A startup running thousands of document-review agents routes every routine step, reading files, extracting fields, calling tools, through Muse Ember, reserving a larger model only for planning, and cuts its per-task inference bill to a fraction of before.

#06 · Open Source

Mistral open-weights Large 4, a trillion-parameter European model

# Le Chonk arrives fully open, built for coders and defensive security.
> the_news

French lab Mistral AI released the open weights of Mistral Large 4, nicknamed Le Chonk, a 1.05-trillion-parameter mixture-of-experts model with roughly 50 billion active parameters, a 1.6-billion-parameter vision encoder and a 1M-token context window. Trained from scratch in Europe on 3,800 NVIDIA Grace Blackwell GPUs, it leapt on agentic benchmarks after three weeks of extra reinforcement learning, hitting 41% on Terminal-Bench 4.0 and matching frontier open systems on the Artificial Analysis Coding Agent Index while using about 47% fewer tokens. Weights are on Hugging Face for self-hosting with vLLM or SGLang, and Microsoft Foundry is the first cloud to offer it.

> why_it_matters

A fully open, European-trained frontier model lets organizations fine-tune, self-host and keep data in their own infrastructure, a direct answer to sovereignty and vendor-lock-in concerns, and keeps competitive pressure on closed labs.

> real_life_use_case

A bank that cannot send sensitive data to a hosted API downloads Mistral Large 4, fine-tunes it on its own documents and runs it entirely on-premises, getting frontier coding and reasoning without any customer data leaving the building.

#07 · Funding

Mercor raises $750M at a $20B valuation to evaluate AI agents

# The AI training-data company doubles its value as agent evaluation goes enterprise.
> the_news

Mercor announced a $750 million Series D on November 10 at a $20 billion valuation, double its $10 billion mark from October 2025. The company supplies the human expertise used to train and evaluate AI models, drawing on a network of more than 5 million domain experts. Most of that work has been for frontier labs; the new money funds a push to help enterprises in every industry measure whether an AI agent can be trusted with a real job, including building more of the enterprise-app training environments it gained by acquiring Deeptune. The round drew Nvidia, Salesforce, Zoom, Benchmark, Felicis and General Catalyst.

> why_it_matters

As companies hand real work to AI agents, they need rigorous ways to prove those agents are reliable. Mercor's jump in valuation signals that evaluation, not just model building, is becoming a core layer of the enterprise AI stack.

> real_life_use_case

Before letting an AI agent process insurance claims, a carrier uses Mercor's domain experts and simulated environments to stress-test it on thousands of realistic cases, getting a trust score and failure map before a single live customer is affected.

#08 · Infrastructure

Rackspace and Together AI sign $380M GB300 inference deal

# A multiyear bet on dedicated NVIDIA clusters for managed AI inference.
> the_news

Rackspace Technology announced on November 10 a definitive agreement with Together AI worth roughly $380 million in anticipated total contract value over several years. Rackspace will stand up and operate dedicated NVIDIA GB300 NVL72 clusters, with Spectrum-X Ethernet networking, liquid cooling, power and storage, for Together AI's managed inference workloads, targeting mid-2027 service. The two also signed a non-binding letter of intent covering potential multi-billion-dollar future capacity and a joint enterprise AI go-to-market. Alongside a $230 million AMD-based deal with Blacksmith, Rackspace now has about $610 million in signed AI infrastructure contracts as it pivots from legacy cloud to enterprise AI.

> why_it_matters

The deal shows demand for specialized, operated GPU infrastructure, not just raw capacity, as inference workloads scale. For legacy cloud players, operating frontier AI clusters is becoming a path to reinvent a shrinking core business.

> real_life_use_case

An enterprise in a regulated industry runs its AI inference on Together AI's platform knowing Rackspace operates the underlying GB300 clusters with the security, governance and compliance controls its auditors require, instead of self-managing scarce GPUs.

#09 · Robotics

Path Robotics raises $300M to scale AI welding robots

# Physical AI for manufacturing lands a Series E at a $2.45B valuation.
> the_news

Path Robotics announced a $300 million Series E on November 10 at a $2.45 billion post-money valuation, led by Georgian with the venture arms of Yamaha Motor and South Korea's Navier joining. The startup automates industrial welding with an AI model called Obsidian, built to power robots equipped with welding torches. Obsidian observes how metal parts change shape during welding and adjusts its workflow in real time, automatically accounting for unexpected part variations, a long-standing obstacle to automating a craft that depends heavily on scarce, skilled human welders.

> why_it_matters

Welding is a high-skill, high-shortage trade, and most parts vary just enough to defeat rigid automation. An AI that adapts to each weld in real time could ease a labor crunch and bring flexible automation to small-batch manufacturing.

> real_life_use_case

A metal-fabrication shop struggling to hire welders deploys Path's robots on short production runs; the Obsidian model adapts to each part's slight variations on the fly, delivering consistent welds without reprogramming for every new job.

#10 · Industry

OpenAI launches Businesses in ChatGPT for small teams

# A dedicated home for companies to run ChatGPT together.
> the_news

OpenAI introduced Businesses in ChatGPT, a new offering giving companies a shared, managed home for ChatGPT inside their organization. The move targets small and mid-sized teams that want collaborative workspaces, shared tools and administrative controls without the overhead of a full enterprise deployment, continuing OpenAI's push to turn ChatGPT from a consumer assistant into everyday business infrastructure. It arrives amid a broader enterprise sprint across the major labs, as OpenAI reported its Q3 revenue climbed roughly 50% quarter over quarter to around $10 billion on renewed demand for recent models and enterprise products.

> why_it_matters

Making ChatGPT easy for whole companies to adopt, with shared workspaces and admin controls, is how OpenAI converts massive consumer familiarity into durable business revenue and fends off rivals courting the same enterprise seats.

> real_life_use_case

A 30-person agency sets up Businesses in ChatGPT so staff share prompts, connected tools and brand context in one managed workspace, while an admin controls access and data settings, instead of everyone paying for disconnected personal accounts.

> source

OpenAI ↗

# get this in your pocket every morning
Join on WhatsApp →
# AI Majlis Daily · Abu Dhabi · Dubai · Sharjah