Anthropic Launches Claude Fable 5.1
Anthropic has released Claude Fable 5.1, its most capable model for demanding coding, research, and long-running agentic work, with major improvements in efficiency and complex task execution.
- Fable 5.1 and Mythos 5.1 share the same underlying model, but Fable adds safeguards that restrict potentially dangerous cyber, biology, and chemistry requests.
- The new model improves across coding, computer use, scientific reasoning, agentic tasks, and professional knowledge work, outperforming Fable 5 across several Anthropic evaluations.
- Anthropic redesigned its safety system so Fable 5.1 can now help identify vulnerabilities directly from source code, while still blocking exploit generation, penetration testing, and binary-based vulnerability scanning.
- Biology safeguards have also become less intrusive, with Anthropic reporting 85% fewer interventions on benign biology requests compared with the safeguards originally deployed with Fable 5.
- For vetted researchers and security organizations, Claude Mythos 5.1 provides fewer restrictions on advanced cyber and life-sciences work, with pricing starting at $10 per million input tokens and $50 per million output tokens.
Anthropic is trying to make its most capable underlying model broadly useful without exposing its highest-risk capabilities to everyone. Fable 5.1 shows how frontier labs may increasingly ship the same core intelligence at different access levels depending on the sensitivity of the task.
Meta Puts Robots to Work in Data Centers
Meta is testing robots inside its massive AI data centers to handle physical maintenance jobs traditionally performed by human technicians, from swapping cables to resetting servers.
- Meta is testing robotic systems from companies including Kinova, ABB, and Watney Robotics.
- The robots are being trained to replace network cables, power-cycle servers, inspect hardware, and transport equipment.
- Some internal estimates suggest automation could eventually handle up to 80% of certain technician workloads inside data centers.
- Meta is already piloting newer robotic systems at facilities in Iowa and Ohio, while simpler robots are used for inventory and moving server racks.
- The technology is still in the testing phase, with challenges around battery life, navigation, speed, and complex hardware tasks.
The AI boom is creating enormous demand not only for GPUs and electricity, but also for people who physically maintain millions of servers. If robots can automate repetitive maintenance, Meta could operate increasingly massive AI data centers with fewer on-site interventions while keeping human technicians focused on more complex repairs.
Dyson Puts AI Inside a $500 Toothbrush
Dyson is entering oral care with CameraJet, a $500 smart toothbrush that uses a built-in camera and machine learning to automatically find and clean gaps between your teeth.
- CameraJet packs a 100,000-pixel macro camera that captures 28 images per second, giving its AI system a real-time view inside your mouth.
- Its Gap Optical Targeting system was developed using more than 470,000 dental images and can detect an interdental gap and trigger a cleaning jet within 100 milliseconds.
- Instead of traditional flossing, a built-in 12.5ml tank fires a targeted conical jet of mouthrinse between teeth while you brush, combining brushing and flossing in one device.
- CameraJet runs on roughly 16 million lines of code, connects through Wi-Fi and Bluetooth, and sends live video, coverage maps, and personalized brushing feedback to the MyDyson app.
- Dyson spent six years developing the device with 661 engineers and filed 38 patents. It costs about $500, with preorders underway and shipping beginning September 8 in initial markets.
Dyson is applying computer vision and machine learning to a category that has changed relatively little in decades. CameraJet turns the toothbrush into a real-time sensing device that can identify where cleaning is needed and act on that information automatically.
Anthropic’s $7B MatX Acquisition Talks Shift
Anthropic reportedly discussed acquiring AI chip startup MatX for around $7 billion, but negotiations have since shifted toward a potential partnership instead.
NVIDIA Forecasts 70% Revenue Growth
NVIDIA forecast 70% fiscal 2028 revenue growth and projected $108 billion in third-quarter sales, fueled by surging AI infrastructure demand.
Cursor Lets Cloud Agents Start Without Repositories
Cursor now lets developers launch Cloud Agents without connecting GitHub first, automatically creating an Origin repository that can be moved to a private repo later.
ChatGPT Connects With Epic Healthcare Records
OpenAI integrated ChatGPT for Healthcare with Epic, letting clinicians securely summarize patient records and access supporting information alongside nine official healthcare data sources.
Runway Unveils Real-Time AI Interfaces
Runway previewed Solaris, an Interface World Model that generates interactive visual interfaces in real time and responds directly to clicks and user actions without first producing code.
Google Adds Gemini Memory and Guided Vision to Android
Google is bringing item-location memory to Gemini and adding Guided Vision, which uses the Gemini Live camera to help blind and low-vision users navigate their surroundings.
AI Agent Security Startup AIR Raises $50M
AIR raised $50 million from investors including Sequoia and Greenoaks to secure enterprise AI agents by continuously inspecting their tools, skills, and interactions.
Someone take Astra’s Wi-Fi away
OpenAI says its upcoming model, Astra, has reached what it calls a “Critical” level of cybersecurity capability, making it the first OpenAI model to reach that threshold.
In testing, Astra was able to find previously unknown security flaws and turn them into working exploits against protected systems.
It also performed much better than GPT-5.6 Sol on internal cybersecurity tests, including finding two new vulnerabilities during one evaluation.
Because of those results, OpenAI delayed parts of Astra’s development while it added stronger safeguards.
These include stricter refusals for harmful cyber requests, more monitoring and systems designed to stop the model if it tries to act outside its allowed limits.
In brief:
- Astra can find and exploit serious security flaws.
- OpenAI has added stronger safeguards and monitoring before release.
- Its most advanced cyber tools will have limited access at launch.
Ah. So it can hack.
OpenAI says Astra is also better at following security restrictions than GPT-5.6 Sol.
In one test, Astra refused 91.5% of harmful cyber requests, compared with 59% for GPT-5.6 Sol.
Astra is still expected to launch, but its most advanced cybersecurity features will initially be limited to a smaller group of testers before access expands.
The brief was in Slack all along
Adobe is bringing more than 70 of its creative and productivity tools directly into Slack, so teams can create and edit content without leaving their chats.
Adobe for Slack is now available for Slack Business+ and Enterprise+ customers.
It includes tools from Photoshop, Premiere, Acrobat, Illustrator, Lightroom, Adobe Express and Firefly.
Users can ask Slackbot what they want to create, and it can use the right Adobe tools to do it.
Slackbot can also pull context from conversations, files, channels and Canvases, so teams can work with information already sitting in Slack.
For example, users can turn project discussions into PDFs, create social media versions of campaign assets, find old Creative Cloud files, edit batches of images, remove backgrounds, resize content, and turn spreadsheet data into visuals.
Here’s what you should know:
- More than 70 Adobe tools are now available through Slackbot.
- Slack content can be used as context for creating and editing work.
- Teams can make PDFs, images, videos and campaign assets without constantly switching apps.
The browser tab industrial complex takes another hit
Teams can then share the results in Slack, ask for edits or create more versions.
For more detailed work, they can still move into Adobe apps such as Photoshop, Premiere, Acrobat and Firefly.
Adobe says the Slack app is part of a wider push to bring its tools into popular AI assistants.
Its tools are also available through ChatGPT, Claude and Copilot, with Gemini support coming soon.
Users can try Adobe for Slack without an Adobe account, but signing in unlocks more tools, generative AI features and Creative Cloud files.
Anthropic open-sources Claude shopping agents that boost cart size by 35%
Anthropic just open-sourced Claude Commerce Agents, a ready-to-use blueprint for building AI shopping assistants. Think of it as a starter kit so you skip weeks of figuring out how to wire an AI agent to a store backend.
There are two agents included:
- Shopping agent: sits inside a customer-facing app, searches products, fills carts, and answers order questions
- Merchant agent: helps store staff manage listings, inventory, and pricing, but every change waits for a human to approve it
- Four vertical demos: retail, travel, telecom, and entertainment, all runnable out of the box
- A Claude Code plugin that scaffolds an agent against your own backend automatically
Real results from pilots: 35% larger carts and shoppers 60% more likely to complete a purchase. To get started, clone the repo, add your API key, and run
python scripts/run_demo.py retail
Qwen upgrades its 2.4T parameter model with 1M token context at $2 per million
Alibaba just pushed an update to Qwen3.8-Max, and the new version is called Qwen3.8-Max-0902. Same giant model underneath, but with extra post-training focused on coding and collaborative work tasks.
Quick recap of what this thing actually is: a 2.4-trillion-parameter model with 95B active parameters and a 1M-token context window. That context window means you can feed it an entire large codebase in one shot.
The upgrade sharpens it for real enterprise work. Here is what it handles well:
- PaperBench score of 93.0, ahead of GPT-5.6 Sol, Claude Fable 5, and Opus 4.8
- OSWorld-Verified score of 86.1, beating GPT-5.6 Sol Max and Claude Fable 5
- Strong vision across satellite imagery, documents, technical drawings, and crowded scenes
Pricing stays the same: $2 per 1M input tokens and $6 per 1M output tokens, plus cache discounts. You can hit it right now via the QwenCloud API.
Google ships Gemini 2.5 Flash with stronger coding and multi-step reasoning at same price
Google just dropped Gemini 3.8 Flash, and the pitch is simple: smarter reasoning, same low price as 3.7 Flash.
So what actually changed? The model now works harder on tough tasks. The performance boost comes from a design choice: it executes extra internal reasoning steps and calls tools repeatedly to reach better answers. Think of it like a teammate who keeps double-checking their work instead of stopping at the first answer.
Here is what it is good at:
- On long-horizon coding, it beats most larger, pricier models at solving complex engineering problems end to end.
- On Terminal-Bench 2.1, an agentic terminal coding test, it scores 89.4%, edging out Claude Opus 5.
- It handles text, image, audio, video, and PDF input with a 1M token context window.
Pricing stays at $0.75 per 1M input tokens and $3.75 per 1M output. If you need lower compute overhead, you can adjust effort settings or stay on 3.7 Flash. You can try it now via the Gemini API and AI Studio.
Runway’s Latest Magic Trick
Runway released Solaris, an Interface World Model that generates apps frame by frame without code. It also released its own benchmark showing Solaris beats Claude Opus 5 at following instructions. Runway wrote the test. Runway won the test. Last year Runway tried the same playbook with Seedance, wrapping a partner model and calling it a new product. The pattern is the pattern.
Solaris is a vision model plus an LLM. The LLM decides what happens. The vision model renders it. The demo is a virtual clothing store. A top language model with MCP could do the same thing. Connect it to a rendering engine, give it tool access, and it generates the same frame-by-frame interactions. The benchmark is the tell. Runway tested Solaris against a model that was never designed to do what Solaris does, on a test Runway wrote itself, and then declared victory. The technology is real. The framing is a magic trick.
Runway built a “real product”, benchmarked it against a model that was not designed for the task, won its own test, and called it a revolution. The technology is real. The framing is a magic trick.
Gemini Quits the Model Race
Google launched Gemini 3.8 Flash, its third Flash in six weeks. Gemini 3.5 Pro was canceled. Gemini 4 is stuck in post-training. HLE-Verified: 54.9%, barely touching the frontier. Price: $0.75 per million input, $3.75 per million output. DeepSeek V4 Flash is cheaper and better. GLM 5.3 Flash costs a fraction. Google walked out of the frontier race and walked into a budget race it was already losing.
Google bet on a middle ground that does not exist. The frontier belongs to Anthropic and OpenAI. The budget lane belongs to DeepSeek and Zhipu. 3.8 Flash is neither the best nor the cheapest. It is the most expensive cheap model on the market. Google did not choose this strategy. Pro collapsed, Flash filled the gap, and the gap turned out to be a dead zone. The market rewards two things: the best and the cheapest. Google is neither.
Google left the frontier race, entered the budget race, and discovered it was already losing both.
The US Nationalized Your Content
The Trump administration’s Justice Department filed court papers backing OpenAI in its copyright fight with the New York Times. The argument: training AI on millions of news articles without permission is fair use, because the public benefits outweigh any competitive harm. The DOJ went further. It said AI models help national security officials analyze real-world facts. Implying, without quite saying it, that your copyright is a threat to American prosperity.
Fair use has four factors. Purpose of the use. Nature of the work. Amount taken. Market impact. The DOJ added a fifth: does the AI help America? If yes, the first four do not matter. The New York Times called it what it is: the government siding with trillion-dollar AI companies against the creators whose work they took. The DOJ’s argument is not legal. It is geopolitical. Copyright is a speed bump on the road to AI dominance, and the government just asked the court to remove it, wrapped in the flag.
The US government just told a court that AI training is more important than copyright, and the national security argument was the velvet glove on the fist.
AI’s 1% Ate the Whole Capital Pool
Mercury surveyed 1,500 US early-stage founders. AI adopters are four times more likely to raise VC. 31% versus 7%. But even the 31% are raising pocket change compared to the top. OpenAI, Anthropic, and the handful of AI companies at the head are absorbing the entire capital pool. The 1,500 founders are not the AI industry. They are the long tail of an industry where the head is a different species.
Every previous tech boom had a winner-take-most effect. SaaS spread capital across a hundred companies. Mobile across dozens. AI is compressing it into a handful. The head companies are raising in billions, building their own data centers, and hiring the talent pool dry. The 1,500 founders in the survey are building real companies. The venture market is treating them like background noise.
AI was supposed to democratize the startup economy. It did the opposite. It concentrated capital more than any industry before it, and the tail is finding out in real time.
Microsoft to Disclose Azure Revenue for First Time
Microsoft announced it will begin reporting Azure quarterly revenue as a standalone segment, ending years of opacity around its cloud business and giving investors unprecedented visibility into the company’s most important growth engine.
Anthropic Open-Sources Claude Commerce Agents
Anthropic released Claude Commerce Agents as an open-source blueprint for building shopping and merchant AI agents, setting a new standard for commercial agent development and lowering the barrier for developers to build transactional AI experiences.
Google Ships Gemini 3.8 Flash with Cyber Edition
Google released Gemini 3.8 Flash alongside a new Cyber variant, promising the model works harder on complex reasoning tasks but may come with higher compute costs, as the AI model race intensifies across price-performance tradeoffs.
Researchers Warn of OpenAI Astra Safety Disaster
Ahead of OpenAI’s Astra release, researchers are sounding alarms about a potential safety race to the bottom as the autonomous AI monitoring system approaches deployment, raising questions about industry-wide safety standards for self-supervising AI.
Trump Administration Backs OpenAI Against NYT
The US Justice Department filed a brief supporting OpenAI in its copyright battle with The New York Times, arguing that restricting AI training on copyrighted works could cause America to fall behind in the global AI race.
OpenAI Sued for ‘Aiding’ Mass Shooting
Dozens of new lawsuits accuse OpenAI of aiding and abetting the Tumbler Ridge mass shooting, testing the legal boundaries of AI product liability and potentially redefining platform responsibility for harmful AI outputs.
Meta May Cut Employee Access to Rival AI Models
Market speculation suggests Meta will block employees from accessing Anthropic and OpenAI models, reflecting growing competitive isolation among AI platforms as the battle for talent and tooling intensifies.
Google Defeats DOJ Bid to Break Up Ad Business
A federal court rejected the DOJ’s proposal to force Google to sell its advertising technology business, preserving the company’s $200 billion ad empire and marking a major win in its ongoing antitrust battles.
Uber Launches Robotaxis in London Before Waymo
Uber partnered with Wayve to launch lidar-free, geofence-free robotaxis in London, beating Waymo to become the first operator of autonomous ride-hailing in the UK capital.
Lyte Raises $165M for Robot Perception
Robotics perception startup Lyte raised $165 million at a $1.6 billion valuation to bring centimeter-accurate 3D perception to industrial robots, highlighting growing investor appetite for embodied AI infrastructure.
Qualcomm Packs Neural Fusion GPU Into Next Chip
Qualcomm detailed its Adreno Neural Fusion GPU for the next Snapdragon flagship, embedding AI inference directly into mobile graphics silicon and blurring the line between traditional compute and on-device intelligence.
Nvidia Now Accounts for 8% of S&P 500
Nvidia’s market cap now represents roughly 8% of the S&P 500’s total value, near its all-time high, as one company’s dominance reshapes the structure of America’s benchmark stock index.
Palantir’s Karp Backs Ukraine Defense Startup
Palantir CEO Alex Karp personally invested in a defense tech startup founded by Ukraine’s ousted defense minister, extending Palantir’s influence deeper into battlefield technology and the military startup ecosystem.