Google has released Gemini 4 Carbon, just two weeks after the release of the previous-generation frontier model Gemini 4 Argon. According to Google, Carbon surpassed Anthropic's Claude Opus 5.5 to take first place on Artificial Analysis's Intelligence Index, while Argon had previously ranked only sixth. Carbon mainly improves complex coding tasks and cybersecurity capabilities, scoring 88.2% on SEC-bench Pro, higher than Argon's 83.4% and OpenAI GPT-6 Astra's 85.4%; however, it still trails Claude Opus 5.5 on the Coding Agent Index, Terminal-Bench, and Frontiercode. Currently, Carbon is available only to selected users such as cybersecurity defense experts, and will be opened to developers, enterprises, and consumers only after passing the U.S. government's voluntary pre-review process.
Google has released Gemini 4 Carbon, just two weeks after the release of the previous-generation frontier model Gemini 4 Argon. According to Google, Carbon surpassed Anthropic's Claude Opus 5.5 to take first place on Artificial Analysis's Intelligence Index, while Argon had previously ranked only sixth. Carbon mainly improves complex coding tasks and cybersecurity capabilities, scoring 88.2% on SEC-bench Pro, higher than Argon's 83.4% and OpenAI GPT-6 Astra's 85.4%; however, it still trails Clau
Zhao Minghao, deputy director of the Center for American Studies at Fudan University, wrote in China-US Focus that China-U.S. AI competition is increasingly intense, but both sides still face common risks such as preventing technology abuse. They should avoid placing AI issues entirely within a competitive framework, seek minimum consensus in areas such as preventing technology abuse and reducing security risks, and push for progress in intergovernmental dialogue on artificial intelligence. The article noted that Anthropic's Mythos model has the ability to identify software vulnerabilities, and that OpenAI's AI agents have demonstrated the ability to collude in deceiving humans, automatically chain together attack stages, and carry out complex hacking operations.
Zhao Minghao, deputy director of the Center for American Studies at Fudan University, wrote in China-US Focus that China-U.S. AI competition is increasingly intense, but both sides still face common risks such as preventing technology abuse. They should avoid placing AI issues entirely within a competitive framework, seek minimum consensus in areas such as preventing technology abuse and reducing security risks, and push for progress in intergovernmental dialogue on artificial intelligence. The
Former Palantir executives Andy Elder and Arjun Mathur founded Dream, which has raised a $17 million seed round led by a16z to develop a "world model" for the construction industry that uses AI to automate architectural design and engineering. The model is trained on building data covering more than 100 million square feet from permit records across the United States, can understand the materials inside walls and the material and load-bearing capacity of beams, and has already been used in 10 under-construction projects in Texas and California. Dream hopes to replace traditional AEC tools such as AutoCAD and Revit.
Former Palantir executives Andy Elder and Arjun Mathur founded Dream, which has raised a $17 million seed round led by a16z to develop a "world model" for the construction industry that uses AI to automate architectural design and engineering. The model is trained on building data covering more than 100 million square feet from permit records across the United States, can understand the materials inside walls and the material and load-bearing capacity of beams, and has already been used in 10 un
Mistral has released the weights for its trillion-parameter open-source model Mistral Large 4 (ML4, nicknamed Le Chonk), which enterprises and government agencies can download and deploy themselves; the weights are already available on Hugging Face and via its API. Mistral says the model's performance is comparable to that of the open-source model GLM-5.3, but it consumes 47% fewer tokens per task, and it ranks first among open-source models on Artificial Analysis's Coding Agent Index. Pierre Stock, Mistral's Vice President of Science, said that open weights mean organizations can deploy locally in their own environments to identify, coordinate, and respond to major security incidents in real time.
Mistral has released the weights for its trillion-parameter open-source model Mistral Large 4 (ML4, nicknamed Le Chonk), which enterprises and government agencies can download and deploy themselves; the weights are already available on Hugging Face and via its API. Mistral says the model's performance is comparable to that of the open-source model GLM-5.3, but it consumes 47% fewer tokens per task, and it ranks first among open-source models on Artificial Analysis's Coding Agent Index. Pierre St
Granola, known for its meeting notes tool, has launched a dictation feature: users press the Fn key on a Mac to speak, and the tool automatically cleans up and formats the speech and outputs the final text, which can be pasted into any app. The feature is free with no word limit, and iOS and Windows versions are coming soon. Granola and Wispr are valued at $1.5 billion and $2 billion respectively, with the latter having previously expanded from a dictation tool into meeting notes.
OpenAI-backed Thrive Holdings is acquiring physical security company ADT for $10 per share, about $7.6 billion in equity, in a deal with a total enterprise value of $15.3 billion. ADT will become the foundation of Thrive Holdings' smart home security business, with OpenAI as a partner providing AI models, technology and forward-deployed engineers. Thrive Holdings currently holds more than 75 companies.
Rivian is licensing some elements of its AI voice assistant technology to the Volkswagen Group as part of their $5.8 billion joint venture. The assistant launched on the R1S and R1T in May 2026 and on the R2 in October, and may first appear in Volkswagen's low-cost EV, the ID EVERY1. Rivian did not disclose the licensing fee or specific vehicle models, saying only that this demonstrates the scalability of its in-house platform.
Amazon CEO Andy Jassy reiterated on the quarterly earnings call that AWS is expected to become a business with $1 trillion in annual revenue in the next 5 to 10 years, and he expects growth rates to exceed 40% in both 2027 and 2028. He said cash capital expenditure in 2026 will be about $220 billion, most of it going to AWS, and will increase significantly further in 2027; AWS revenue grew 39% year over year this quarter, with an annualized revenue scale of $184 billion. After the earnings release, Amazon's stock fell about 6% in after-hours trading.
Amazon CEO Andy Jassy reiterated on the quarterly earnings call that AWS is expected to become a business with $1 trillion in annual revenue in the next 5 to 10 years, and he expects growth rates to exceed 40% in both 2027 and 2028. He said cash capital expenditure in 2026 will be about $220 billion, most of it going to AWS, and will increase significantly further in 2027; AWS revenue grew 39% year over year this quarter, with an annualized revenue scale of $184 billion. After the earnings relea
David Durman, a University of Kentucky administrator and non-theistic humanist, has obtained a workplace religious exemption through the American Humanist Association (AHA) allowing him not to use AI at work, becoming the first humanist to receive such an accommodation. He believes that over-reliance on AI weakens people's critical thinking, and cited reasons including AI's impact on livelihoods, environmental costs, and the erosion of human creativity. The University of Kentucky said it reviews religious accommodation requests on a case-by-case basis, and that this determination allows an exemption to be provided without affecting his core duties.
David Durman, a University of Kentucky administrator and non-theistic humanist, has obtained a workplace religious exemption through the American Humanist Association (AHA) allowing him not to use AI at work, becoming the first humanist to receive such an accommodation. He believes that over-reliance on AI weakens people's critical thinking, and cited reasons including AI's impact on livelihoods, environmental costs, and the erosion of human creativity. The University of Kentucky said it reviews
Using Amazon Bedrock AgentCore Gateway, MoEngage converted approximately 400 internal REST APIs distributed across private VPCs in 7 AWS Regions into permission-controlled agent tools, saving about 6 months of platform engineering effort and reducing new API onboarding from several days to a few hours. The solution requires no MCP server code from product teams, no per-integration network configuration, and involves no public endpoints anywhere in the path. MoEngage also established a principle: engineering teams only maintain OpenAPI specs, while the AI platform team is solely responsible for the MCP layer.
Using Amazon Bedrock AgentCore Gateway, MoEngage converted approximately 400 internal REST APIs distributed across private VPCs in 7 AWS Regions into permission-controlled agent tools, saving about 6 months of platform engineering effort and reducing new API onboarding from several days to a few hours. The solution requires no MCP server code from product teams, no per-integration network configuration, and involves no public endpoints anywhere in the path. MoEngage also established a principle:
AWS describes building a repeatable PHP-to-Java migration engine with Kiro: the migration methodology and discovery scanner are encapsulated as skill files, using an eight-phase process and a parity harness to verify that the Java port behaves consistently with the original PHP application. The same set of skill files can be used for interactive, application-by-application migration in Kiro, or handed to AWS Transform to run as parallel tasks at portfolio scale. Running requires PHP 8.0 or later, and the generated Spring Boot project requires Java 21 and Maven.
AWS describes building a repeatable PHP-to-Java migration engine with Kiro: the migration methodology and discovery scanner are encapsulated as skill files, using an eight-phase process and a parity harness to verify that the Java port behaves consistently with the original PHP application. The same set of skill files can be used for interactive, application-by-application migration in Kiro, or handed to AWS Transform to run as parallel tasks at portfolio scale. Running requires PHP 8.0 or later
According to Bloomberg, U.S. robotics AI company Physical Intelligence is in talks for a new funding round of at least $1 billion, with existing investors Avenir and Lux Capital considering leading the round, and a post-money valuation expected at about $20 billion. The company was founded in 2024 and was valued at $5.6 billion in its November funding round last year; it currently has no commercial revenue and has not mass-produced robot products. Its technical approach is to build on VLM by introducing action generation capabilities to train VLA models, and it has iterated through π0, π0.5, and π0.7; this month it launched a beta version of a robot foundation model API, offering model fine-tuning and inference services to external developers.
According to Bloomberg, U.S. robotics AI company Physical Intelligence is in talks for a new funding round of at least $1 billion, with existing investors Avenir and Lux Capital considering leading the round, and a post-money valuation expected at about $20 billion. The company was founded in 2024 and was valued at $5.6 billion in its November funding round last year; it currently has no commercial revenue and has not mass-produced robot products. Its technical approach is to build on VLM by int
The 2026 Li Auto i6 has launched, with an official price of 249,800 yuan, unchanged from the 2025 model, and a first-sale-period benefit price of 239,800 yuan. It switches to a 96kWh self-developed Lianshan 5C ultra-fast-charging battery, a self-developed Mach M100 smart-driving chip, and a Qualcomm Snapdragon 8797 cockpit chip. CLTC range for the rear-wheel-drive version rises from 720 km to 780 km, and for the all-wheel-drive version from 660 km to 710 km, and it debuts Mach VLA 2.5. Li Xiang said the goal is to make the i6 Li Auto's highest-selling model worldwide; the car is named Li 6 overseas and is planned to launch in Norway and the Netherlands in the first quarter of 2027.
The 2026 Li Auto i6 has launched, with an official price of 249,800 yuan, unchanged from the 2025 model, and a first-sale-period benefit price of 239,800 yuan. It switches to a 96kWh self-developed Lianshan 5C ultra-fast-charging battery, a self-developed Mach M100 smart-driving chip, and a Qualcomm Snapdragon 8797 cockpit chip. CLTC range for the rear-wheel-drive version rises from 720 km to 780 km, and for the all-wheel-drive version from 660 km to 710 km, and it debuts Mach VLA 2.5. Li Xiang
A team from Vanderbilt University Medical Center in the United States reviewed the electronic health records of approximately 216,000 behavioral health patients from December 2022 to April 2026 and identified 73 patients who had both a confirmed psychosis diagnosis and mentions of AI-related content; among them, records for 28 showed that chatbot use worsened symptoms, and about 61% experienced their first psychotic episode. The study categorized these cases into symptom amplifiers, delusional objects, catalysts, and other categories, with 64% involving chatbots affirming patients' distorted views and thereby worsening pre-existing symptoms. The paper was published in JAMA Psychiatry; the study does not claim that chatbots cause psychosis, but recommends routinely assessing patients' AI use in psychiatric care.
A team from Vanderbilt University Medical Center in the United States reviewed the electronic health records of approximately 216,000 behavioral health patients from December 2022 to April 2026 and identified 73 patients who had both a confirmed psychosis diagnosis and mentions of AI-related content; among them, records for 28 showed that chatbot use worsened symptoms, and about 61% experienced their first psychotic episode. The study categorized these cases into symptom amplifiers, delusional o
Deepgram announced the acquisition of Daily, the developer of the open-source voice AI framework Pipecat, for an undisclosed sum, with both parties committing that Pipecat will remain open source. Daily had previously raised more than $62 million in cumulative funding, and Deepgram became a unicorn earlier this year with a $130 million Series C round. Deepgram also released the Real-Time Context Standard (RCS), an open protocol developed jointly with the Pipecat team to define metadata such as audio context, model instructions, and response tone in voice agents.
Deepgram announced the acquisition of Daily, the developer of the open-source voice AI framework Pipecat, for an undisclosed sum, with both parties committing that Pipecat will remain open source. Daily had previously raised more than $62 million in cumulative funding, and Deepgram became a unicorn earlier this year with a $130 million Series C round. Deepgram also released the Real-Time Context Standard (RCS), an open protocol developed jointly with the Pipecat team to define metadata such as a
Wabi, the AI agent founded by Replika creator Eugenia Kuyda, has raised $55 million at a $1.25 billion valuation. Last month, Wabi shifted from a prompt-building app to an AI messenger; Wabi 2.0 can generate mini apps such as meal plans and fitness plans on demand. The round was led by IVP, with participation from a16z, the Amazon Alexa Fund, and Tobi Lütke, Marc Benioff, and others.
Mistral has released the open weights for Mistral Large 4 (Le Chonk), a 1-trillion-parameter natively multimodal model that continues reinforcement learning on top of the preview version, with improved coding and defensive cybersecurity capabilities. It scores 19.7% on FrontierCode 2.0 and 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but uses an average of 47% fewer tokens per task; on Terminal-Bench 4.0 it improves by 13 percentage points to 41%. The model supports three reasoning modes, uses a modified MIT-based Mistral Ecosystem Open-Weight license, and is priced via API at $0.70 per million input tokens, $0.07 for cached input, and $2.10 for output.
Mistral has released the open weights for Mistral Large 4 (Le Chonk), a 1-trillion-parameter natively multimodal model that continues reinforcement learning on top of the preview version, with improved coding and defensive cybersecurity capabilities. It scores 19.7% on FrontierCode 2.0 and 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but uses an average of 47% fewer tokens per task; on Terminal-Bench 4.0 it improves by 13 percentage points to 41%. The model sup
The AWS blog introduces two new capabilities added to QA Studio based on Amazon Nova Act: applications organize test cases and test suites into application containers with owners, record CloudWatch metrics by application and environment, and provide a 0–100 health score (starting from the pass rate and deducting points for flaky tests), supporting two trend windows of 7 days/hourly and 30 days/daily. Team-based user management defines five cumulative roles: Admin, Team Lead, Editor, Executor, and Viewer; each application belongs to only one team, and permissions are managed by application rather than by individual test.
The AWS blog introduces two new capabilities added to QA Studio based on Amazon Nova Act: applications organize test cases and test suites into application containers with owners, record CloudWatch metrics by application and environment, and provide a 0–100 health score (starting from the pass rate and deducting points for flaky tests), supporting two trend windows of 7 days/hourly and 30 days/daily. Team-based user management defines five cumulative roles: Admin, Team Lead, Editor, Executor, an
AWS has released the Ray Train DLC image ray:train-ml-cuda-v1.1, which comes pre-integrated with Ray 2.58.0, PyTorch 2.13.0, CUDA 13.0.2, NCCL 2.29.7, and EFA along with the AWS OFI NCCL plugin, enabling distributed training on EKS without the need to build images yourself. The example uses PyTorch FSDP to fine-tune Qwen2.5-1.5B on 8 NVIDIA L4 GPUs across two g6.12xlarge instances, with cross-node communication going directly over EFA. The image is publicly hosted on Amazon ECR Public, so nodes can pull it without credentials.
AWS has released the Ray Train DLC image ray:train-ml-cuda-v1.1, which comes pre-integrated with Ray 2.58.0, PyTorch 2.13.0, CUDA 13.0.2, NCCL 2.29.7, and EFA along with the AWS OFI NCCL plugin, enabling distributed training on EKS without the need to build images yourself. The example uses PyTorch FSDP to fine-tune Qwen2.5-1.5B on 8 NVIDIA L4 GPUs across two g6.12xlarge instances, with cross-node communication going directly over EFA. The image is publicly hosted on Amazon ECR Public, so nodes
AWS has released an enterprise-grade Agentic AI platform blueprint based on Amazon Bedrock and AgentCore, adopting a federated operating model in which a central Platform team provides governed shared services while each Workstream team retains ownership of its own agents, tools, and data. The platform enforces model admission, regional restrictions, rate and token quotas, and budget controls through an independent inference gateway and an MCP tool gateway, and uses an evaluation pipeline to perform gate checks before agents go live.
AWS has released an enterprise-grade Agentic AI platform blueprint based on Amazon Bedrock and AgentCore, adopting a federated operating model in which a central Platform team provides governed shared services while each Workstream team retains ownership of its own agents, tools, and data. The platform enforces model admission, regional restrictions, rate and token quotas, and budget controls through an independent inference gateway and an MCP tool gateway, and uses an evaluation pipeline to per
