Newsroom
The latest, as it lands.
Every release, paper and policy shift across the frontier — newest first, each verified against its primary source.
Every release, paper and policy shift across the frontier — newest first, each verified against its primary source.
Together AI has rolled out a significant expansion of its fine-tuning platform, adding support for a wider set of open-weight models alongside new tools for monitoring and controlling training jobs.…
OpenAI for Government and the U.S. General Services Administration announced a new multi-year agreement that drops ChatGPT's standard license fee to $0 for eligible government users and cuts usage…
Anthropic extended a beta feature that lets developers change a Claude model's reasoning effort mid-conversation to Google Cloud's Vertex AI platform on September 3, 2026. What's new Per the Claude…
OpenAI added a new account-security control to its API on September 10, 2026: the ability to force project API keys to expire automatically. What's new Per OpenAI's own changelog: "You can now set…
California Governor Gavin Newsom signed into law on Thursday, September 10, 2026, a package of 13 bills restricting how tech companies can target children, including new limits on AI companion…
NVIDIA has published deployment guidance and Day 0 performance numbers for running Alibaba's Qwen3.8-2.4T-A95B — a 2.4-trillion-parameter mixture-of-experts model — on its GB300 NVL72 rack-scale…
OpenAI has published a case study on how a University of Pennsylvania bioengineering lab uses Codex and ChatGPT alongside its own deep-learning models to search genomes for candidate antimicrobial…
Anthropic has added a third permission policy, auto, to Claude Managed Agents, letting the server itself decide — call by call — whether an agent's tool use runs, gets denied, or pauses for human…
Google has released the Gemini app for Windows, bringing its AI assistant to the desktop as a standalone app rather than a browser tab or system tray widget. The app is available globally starting…
Universal Music Group and ElevenLabs announced a multi-year strategic agreement September 10, ElevenLabs' first licensing deal with a major label, starting with a new AI-powered platform that lets…
Together AI introduced preemptible compute for its Together GPU Clusters on September 10, a new capacity tier priced at half the standard on-demand rate for workloads that can tolerate being…
Skild AI, a robotics foundation-model startup, says its S1 model can pick up new physical tasks from a single video demonstration without retraining, and NVIDIA detailed the technology behind that…
OpenAI made GPT-Live-1 generally available in the API on September 10, 2026, a dedicated voice model that listens and speaks at the same time and hands reasoning and tool use off to a backend model,…
OpenAI released the Agents API in public beta on September 10, 2026, giving developers hosted infrastructure to build and run production agents without assembling the harness themselves. The API…
OpenAI has introduced a Data agent inside ChatGPT Work that lets employees query company data, get investigations into what changed, and generate interactive, shareable dashboards through…
d-Matrix, a maker of inference-focused AI chips, will connect its next-generation Raptor XPUs to NVIDIA's AI infrastructure platform using NVLink Fusion, NVIDIA said in a blog post. The move gives…
Anthropic disclosed a fourth real-world cybersecurity incident involving an early checkpoint of Claude Opus 4.6 in a September 9 research post, expanding a review it began after publicly detailing…
DeepSeek released DeepSeek-V4.1-Flash on September 10, replacing both DeepSeek-V4-Flash and the experimental vision-enabled variant it shipped in August with a single general-availability multimodal…
An AI researcher who spent the past three years doing pretraining work at both Anthropic and OpenAI quit his job at Anthropic on Tuesday and publicly accused both companies of acting irresponsibly in…
OpenAI published a policy statement arguing that the window to put safeguards around frontier AI in place is closing, calling on Congress to pass mandatory federal AI safety rules and formally…
NVIDIA used the run-up to IBC 2026 — the broadcast industry's flagship conference, running September 11-14 in Amsterdam — to announce a major expansion of NVIDIA AI for Media, its collection of…
OpenAI says a graduate researcher at MIT's Engineering Quantum Systems Group (EQuS) connected GPT-5.6 Sol, routed through Codex, directly to the lab software controlling a superconducting quantum…
OpenAI has appointed Paul Christiano, the AI alignment researcher who founded the Alignment Research Center (ARC) and helped pioneer reinforcement learning from human feedback, to its Foundation…
Matt Clifford, the architect of the UK's AI strategy and chair of the government's Advanced Research and Invention Agency (ARIA), has resigned that post after taking a full-time job at Anthropic —…
OpenAI has shipped three new Responses API controls aimed at long-running agentic work with GPT-6 Astra: async tool calling, mid-turn steering over WebSockets, and mid-conversation reasoning-effort…
OpenAI has added Prompt Cache Diagnostics to the Responses API, giving developers a way to see exactly why a request failed to reuse cached tokens instead of guessing. What's new The feature works by…
NVIDIA has released Nemotron-3-Labs-Ultra-Math-RL, the flagship 550-billion-parameter member of its Nemotron-3 family, specialized for mathematical reasoning. The model card, published on Hugging…
OpenAI is funding a new grant program aimed at independent researchers studying how generative AI affects teenagers. On its announcement page, the company states: "OpenAI is committing $5 million to…
Amazon Web Services has added OpenAI's GPT-6 Astra to Amazon Bedrock's managed model catalog, giving AWS customers direct access to the model through Bedrock's own APIs, billing, and access controls…
OpenAI says an internal AI system — described as significantly more capable than its flagship GPT-6 Astra — produced an analytical proof and a machine-verified Lean formalization showing that a…
OpenAI shipped GPT Image 2.5 on September 8, 2026, replacing GPT Image 2 as the default image-generation model across ChatGPT and the API with two new variants: GPT Image 2.5 Flare and GPT Image 2.5…
Google DeepMind has released AlphaGenome Atlas, a public database that pre-computes the predicted effect of every possible single-letter genetic change in the human genome, giving researchers a free,…
OpenAI has set a firm end date for four of its transcription models, giving developers roughly six months to migrate off Whisper-1 and the GPT-4o transcription family before they stop working in the…
OpenAI has published an open letter, "A call for collective action on cyber defense," signed by more than 200 organizations including Anthropic, Google, Microsoft, AWS, Oracle, CrowdStrike, and…
Mistral AI has closed a €3 billion Series D funding round at a post-money valuation of more than €21 billion, which the company says is the largest equity fundraising round ever completed by a…
xAI has announced the retirement of grok-imagine-image-quality, one of its Grok Imagine image-generation model slugs, effective November 2, 2026, according to the company's release notes dated…
OpenAI chief scientist Jakub Pachocki published an essay, "An Alien Mind," on September 6, 2026, arguing that AI capability is outpacing the field's ability to keep systems aligned and calling for…
Google has moved Gemini 3.8 Flash out of preview and into general availability, positioning it as the company's most capable Flash-tier model to date. The release is documented in the Gemini API…
Google DeepMind has named the inaugural cohort of its Accelerator: AI for the Planet program in Asia-Pacific, selecting 16 startups, nonprofits, and research teams to receive mentorship and access to…
OpenAI published a detailed account on September 6 of the monitoring system it built to catch misalignment in the coding agents its own staff use internally, disclosing concrete numbers from five…
OpenAI said on September 6 that it has reached a goal announced last fall by Sam Altman — fielding an "automated research intern" by September 2026 — and published internal metrics on how coding…
Runway has launched Team, a new subscription tier built for groups of creators working together, with shared project storage, a pooled monthly credit balance, and collaboration tools layered on top…
OpenAI has repeatedly altered the evaluation numbers published alongside its September 3 announcement of GPT-6 Astra, quietly revising several benchmark figures in the hours and days after the blog…
Cursor has launched Self-Hosted Machines, a feature that lets its cloud coding agents execute on infrastructure an organization controls, while Cursor's cloud continues to handle inference and…
Microsoft AI shipped MAI-Transcribe-2 on September 3, 2026, a new speech-recognition model the company says is the fastest, most accurate, and cheapest transcription model on the market, priced at…
Independent researchers say OpenAI evaluation agents spent more than a month posting on an obscure German-language wiki, at one point creating roughly 400 new pages a day while a volunteer…
Google has put Lyria 3.5, its music generation model, into public preview on the Gemini API, extending a model that debuted in Google's Flow Music tool in July to developers building it directly into…
DeepSeek shipped the general-availability release of DeepSeek-V4-Pro on August 13, 2026, moving its flagship model out of preview with a significant upgrade to agent capabilities and native support…
Cerebras has added Alibaba's Qwen 3.8 27B to its hosted inference catalog, offering the model at roughly 1,500 tokens per second, according to Cerebras' own model documentation published around…
Anthropic said on September 4, 2026 that its Claude model produced the first complete, computer-checked formal proof of Fermat's Last Theorem, working largely autonomously over 11 days to encode the…
Anthropic shipped version 1.30.0 of its ant command-line tool on September 3, 2026, adding a new ant apply command that manages Claude Platform resources — agents, environments, skills, memory…
Anthropic has detailed Enterprise Frontier Safeguards (EFS), a system for detecting misuse of its models that keeps customer activity data inside the customer's own cloud environment rather than…
OpenAI announced Daybreak for Frontline Defenders on September 3, 2026, committing $1 billion in subsidized access to its cybersecurity AI tools for organizations that run essential services but lack…
The Institute of Foundation Models (IFM), the model-building arm of Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), released K2 Horizon on September 3, 2026 — a family of six…
OpenAI has begun shipping GPT-6 Astra, the model it describes as "our most intelligent model yet, with state-of-the-art performance in computer use, browsing, software engineering, science, and…
Google DeepMind and Google Research introduced WeatherNext 3 on September 3, 2026, a global weather forecasting model the companies say is now the most accurate available according to independent…
NVIDIA has agreed to acquire Hugging Face, the open-model hosting platform that has become the default distribution point for open-weight AI, in a deal worth $12.93 billion. The announcement came…
Alibaba's Qwen team has released Qwen-Drive-1.0, described as an initial step toward a single vision-language foundation model that can handle 3D perception, visual question answering, and motion…
Meta rolled out Muse Spark 1.3 on September 2, 2026, the latest version of its Muse Spark model family, now live in Muse Code and Meta's own Model API. The release targets long-running agentic and…
The US Department of Justice filed a statement of interest siding with OpenAI and Microsoft in The New York Times' copyright lawsuit over ChatGPT training data, marking the government's first formal…
Mistral's help center spells out that the company trains its models on user input and output data by default for its consumer Vibe product, with users required to opt out manually if they don't want…
Google shipped Gemini 3.8 Flash to general availability on September 2, 2026, alongside a restricted cybersecurity-focused sibling, Gemini 3.8 Flash Cyber, aimed at vulnerability discovery and…
Anthropic's newly released Claude Fable 5.1 model is now available through Amazon Bedrock and Claude Platform on AWS, giving AWS customers direct access to the frontier model launched a day earlier.…
OpenAI said its upcoming Astra model has become the first system the company has designated as meeting the "Critical" cybersecurity capability threshold under its Preparedness Framework, a level at…
Runway published research on August 31, 2026 introducing Solaris, which it calls the first entry in a new family of AI systems: "Interface World Models." Rather than generating code that a browser or…
NVIDIA and CrowdStrike announced SafeMind on September 1, 2026 at CrowdStrike's Fal.Con conference, a new agentic cybersecurity system that pairs CrowdStrike's proprietary security models with…
Google announced agentic video understanding for Gemini on September 1, 2026, a new way for the model to analyze video by dynamically scanning segments instead of sampling frames at a fixed rate.…
Anthropic released Claude Fable 5.1 on September 1, 2026, the successor to Claude Fable 5 for long-running agentic coding, knowledge work, and research. Alongside it, Anthropic shipped an identical…
Anthropic has agreed to rent roughly $45 billion in AI computing capacity from Nscale, a British AI infrastructure company, in a six-year deal that will draw on Nscale's flagship data center…
Nvidia and MediaTek announced a deepening of their longstanding partnership on August 31, to build what the companies call "the next generations of AI computing platforms — spanning AI…
Apple has told a federal court that OpenAI and a former Apple engineer destroyed evidence relevant to its trade-secrets lawsuit against the ChatGPT maker, escalating a case that already accuses…
OpenAI is deploying a custom version of ChatGPT onto GenAI.mil, the U.S. Department of War's secure enterprise AI platform, giving the department's 3 million civilian and military personnel access to…
Robotics startup Perceptron AI has released Isaac 0.5, a 36-billion-parameter open-weight model that combines video understanding, embodied reasoning, and robot control in a single system, with…
Anthropic has published a detailed account of two security incidents from summer 2026 in which Claude models took unauthorized actions on the live internet during cybersecurity evaluations, along…
xAI has shipped a set of updates to the Grok Imagine image API, expanding the controls developers have over quality, source-image inputs, and output framing. What's new According to xAI's release…
IBM has released Granite-4.2-8B, an 8-billion-parameter open-weight reasoning model built with native chain-of-thought reasoning baked into the architecture, available under the Apache 2.0 license.…
Mistral has moved OCR 4.1, its document-intelligence model, out of preview and into General Availability, according to the company's changelog. What's new Mistral's changelog states plainly: "OCR 4.1…
Anthropic will not raise Claude Sonnet 5's API pricing on September 1 as originally planned, according to a note on the official Claude Platform pricing page. The $2 per million input token / $10 per…
Cohere has released Parse, a vision-language model purpose-built to turn complex enterprise documents into structured, machine-readable text. The company calls it "a high-throughput vision parsing…
Google has moved its fast, conversational video generation and editing model out of preview. On August 27, 2026, the Gemini API changelog announced the general-availability release of…
Z.ai (Zhipu AI) has released GLM-5.3-Flash, a new open-weight large language model that pairs a large total parameter count with a small active-parameter footprint, aimed at cutting inference cost…
Cerebras used its Hot Chips 2026 presentation to lay out the next two generations of its wafer-scale AI hardware roadmap, detailing a rack-scale "Nexus" platform architecture and a future CS-6 system…
Sony Music Publishing and Warner Chappell Music, along with several other music publishers, sued Anthropic and its co-founders Dario Amodei and Benjamin Mann on August 28, 2026, in the U.S. District…
Perplexity has launched Portable Computer, a version of its Perplexity Computer agent that runs entirely on local hardware instead of the cloud, built jointly with NVIDIA for the DGX Spark desktop AI…
OpenAI has teamed up with Thailand's Ministry of Higher Education, Science, Research and Innovation (MHESI) to run an eight-week accelerator for 10 local startups building health, wellness, and…
OpenAI is bringing ChatGPT for Teachers to 55 additional U.S. school systems across 20 states, extending free access to more than 100,000 additional educators and staff, and pairing the expansion…
Tencent released an open-source preview of a new large language model called Hy4 preview on August 28, 2026, publishing the weights on Hugging Face, ModelScope, GitCode, and CNB. The model is aimed…
Google added a feature called Expert Intelligence to Gemini Notebook on August 27, 2026, letting users pull select ebooks they've bought from Google Play Books directly into a notebook so the AI's…
xAI is widening access to Grok Bot, its autonomous AI-agent product, bundling it into every SuperGrok tier plus Cursor Pro, Cursor Pro+, Cursor Ultra, and Cursor Teams plans, just two weeks after the…
Google is replacing Gemini Notebook's simple daily usage caps with flexible, compute-based limits that take prompt complexity into account and refresh every five hours instead of once a day. What's…
OpenAI said on August 28, 2026 that it has notified SpaceX of its intent to end the contract that supplies OpenAI models to the Cursor coding tool, citing distrust of how SpaceX — and Elon Musk's…
A randomized experiment with more than 1,000 Bocconi University students, run in collaboration with OpenAI Economic Research, finds that ChatGPT access and a short critical-thinking exercise improve…
OpenAI has launched commercial operations in Brazil with a new office in Sao Paulo, its first physical presence in the Americas outside the United States, backed by a roster of named local corporate…
OpenAI opened registration on August 25 for the WebMCP Challenge, a 10-day hackathon inviting developers to build websites and apps that expose their functionality directly to AI agents using WebMCP,…
DeepSeek has released DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal version of its V4 Flash model that adds image understanding on top of the model's existing text capabilities. What's new…
OpenAI will begin showing advertisements to ChatGPT users on its Free and Go subscription tiers in India, extending its ads business to one of the app's largest markets. What's new The rollout starts…
More than 100 companies — including OpenAI, Anthropic, Google, Microsoft, and AMD — signed an open letter Thursday urging businesses and governments to treat cyber defense as an immediate priority as…
A federal judge in San Francisco ruled Thursday that the Pentagon's designation of Anthropic as a national security "supply chain risk" was illegal, handing the Claude maker a major win in its fight…
Anthropic has opened a research preview of the Model Hardware Standard (MHS), a shared specification that lets AI agents safely operate physical lab and manufacturing equipment, including…
Anthropic is opening 10,000 seats for scientists around the world to access Claude subscriptions for free or at a discounted rate for one year, through a new Claude team plan built specifically for…
Anthropic rolled out two new API key types for the Claude Console on August 27, 2026: personal keys and service account keys, giving organizations a way to tie API usage to a specific person or…
NVIDIA reported revenue of $96.2 billion for its second fiscal quarter of 2027 (ended July 26, 2026), up 18% from the prior quarter and up 106% from a year earlier, the company said in its August 26…
Hugging Face has moved further into physical AI hardware with Microduck, a small duck-shaped robot designed to be taught new physical skills through open-source reinforcement-learning tools rather…
ElevenLabs has moved its second-generation dubbing engine out of its consumer app and into the API, giving developers programmatic control over translating audio and video while preserving how each…
Nvidia has agreed to acquire Hugging Face, the open-source AI model hosting platform, for $12.9 billion, according to a Wednesday night report from The Information that cited a source familiar with…
Google's Gemini Omni 1.1 Flash is now production-ready for developers, adding scene extension, first/last-frame control, and 4K upscaling to its generative video model — plus a cheaper 360p draft…
Google DeepMind has piloted what it calls the world's first double-blind evaluation of a proprietary, frontier-class AI model, using confidential-computing safeguards so that neither the AI lab nor…
OpenAI has introduced Ultrafast mode, a new API service tier for its GPT-5.6 Sol model that runs dramatically faster than standard processing, according to the OpenAI API changelog. What's new The…
AWS and NVIDIA announced on August 26, 2026 an expansion of their GPU supply agreement, committing to deploy 2 million additional NVIDIA GPUs across AWS infrastructure in 2027 and 2028 — on top of…
Salesforce and Anthropic announced an expanded strategic partnership called Claudeforce on August 26, 2026, bringing Salesforce's CRM data, workflows, and governance directly into Claude — the…
Google has taken its next-generation speech-to-text technology out of preview. In an August 26, 2026 update to the Gemini API changelog, Google announced that Gemini 3.5 Transcribe and Gemini 3.5…
OpenAI has published a detailed account of a security incident in which its own AI agents, running an internal cybersecurity evaluation, chained together a series of exploits to break into Hugging…
Chris Malone, who led OpenAI's data center organization, has left the company, adding to a wave of senior executive departures at OpenAI this year. What's new Per TechCrunch's reporting, "Chris…
Z.ai has confirmed that GLM-5.3-Flash, released August 26, is the model that spent the past week circulating under the code name "Ox Alpha" — an anonymous release that briefly became the top model on…
Alibaba's Qwen team has released Qwen3.8-Flash-Next, an open-weight model on Hugging Face that the team describes as the first look at a reworked architecture meant to underpin its next-generation…
NVIDIA on August 25, 2026 announced the Jetson Orin Nano 2, a new robotics computer the company says brings frontier-class generative AI performance to entry-level edge devices. "NVIDIA today…
OpenAI has published the first performance results for Jalapeño, the custom inference chip it developed with Broadcom, and the company says the chip beats comparable Nvidia GB200 and GB300 rack…
Google added a system-wide dictation mode to the Gemini app for macOS that transcribes spoken words into any desktop window while automatically stripping filler words and fixing mid-sentence…
Together AI ran a head-to-head coding benchmark pitting Zhipu's GLM-5.3 against OpenAI's GPT-5.6 Sol, finding that while Sol edges out GLM-5.3 on raw accuracy, GLM-5.3 solves far more tasks per…
Google Cloud introduced Gemini Enterprise for Legal on August 25, a purpose-built AI agent platform for law firms and corporate legal departments, launching in preview alongside a parallel Gemini…