Coverage hub
Models
New model releases, upgrades, and capability jumps — frontier labs and open weights alike.
OpenAI Releases Mathematical Results From Internal Frontier Model
Results include Lean formalizations and compute metrics, guided by the Institute for Advanced Study's advisory group.
ModelsGoogle DeepMind Releases EmbeddingGemma 2 for On-Device Multimodal Embeddings
The 740M-parameter model supports text, code, images, audio, and video with a 4x larger context window.
ModelsTII Launches Falcon-Emirati-7B for Dialect-Specific Arabic Language Tasks
The 7B model scores 84.83% on the Alyah benchmark and demonstrates significantly higher dialect fidelity than larger multilingual competitors.
ModelsLiquid AI Releases Open Multimodal Decision Models for Edge Devices
The new d1-3B model answers queries in 16 milliseconds on NVIDIA Jetson AGX Thor hardware.
ModelsAi2 Releases Olmo-Core 3 to Scale MoE Training Infrastructure
The new framework achieves 2.7x throughput over previous implementations on 47B parameter models.
ModelsAmazon Releases Open Source Decision Model Inspired by TypeSafe’s Jev
Strands Decider 2B is built on Qwen3.5-2B and offers low-latency, calibrated choices for agentic workflows.
ModelsGoogle Releases Gemini 4 Argon for Complex Workflows and Cybersecurity
The new model features a 1 million token output limit and is initially available to trusted cyber defenders via the Fairwind Program.
ModelsMusk’s AI Chatbot Grok Reportedly Encouraged Trump to Capture Venezuela’s President
Time reports that President Trump consulted Grok on public sentiment in Venezuela prior to the January invasion.
ModelsElon Musk’s AI-Powered Grokipedia Resumes Updates After Months-Long Pause
The platform’s live updates site shows recent changes, including fact-checks on pages for Barack Obama and Elon Musk.
ModelsOpenAI Releases GPT-6.1 Sol With Near-Astra Performance at Lower Cost
The new model matches GPT-6 Astra on agentic coding tasks while reducing standard input and output token prices to one-fifth of Astra's.
ModelsNVIDIA Releases Open Kumo Tabular Model for Zero-Shot Prediction
The open foundation model ranks first on TabArena with an ELO of 1950 while running 17 times faster than LimiX-2.
ModelsGoogle to Replace Gemini’s Gems With ‘Skills’ in November Update
Custom assistants will automatically migrate to a new format requiring users to type a forward slash to invoke them.
ModelsPeak XV Raises Surge Seed Investment Ceiling to $5M With New 18-Startup Cohort
The new batch, Surge 12, includes AI, robotics, and space startups, with at least three having raised outside funding before joining.
ModelsAnthropic’s Claude Sonnet 5.5 Nearly Matches Opus 5.5 While Costing Less
The new mid-tier model reportedly reduces per-task costs by up to 30 percent compared to its flagship predecessor.
ModelsBasis Completes Tax Workbook 2X Faster With GPT-6 Astra
The accounting AI agent reports a 20% improvement in internal evaluation scores compared to previous models.
ModelsNvidia Releases Free 100-Million-Parameter Speaker Identification Model
The new open-weight model is designed to identify up to eight distinct speakers in real time.
ModelsResearchers Plug GPT-6 Astra Directly Into a Robot to Clean Unfamiliar Kitchen
The experiment demonstrates direct large language model integration for robotic task execution in novel environments.
ModelsWhite House Orders Pre-Review of AI Models Before Sharing With UK AI Safety Institute
The directive requires US government assessment of new models from OpenAI and Anthropic prior to any international testing arrangements.
ModelsAnthropic Says Claude Discovered New Enzyme System; Researchers Call It Routine Mining
The discovery involves a novel enzyme system, but CRISPR experts argue the methodology is standard genome mining.
ModelsDeepMind Chief Prioritizes Gemini 4 Launch Over AGI Research
New leadership shifts focus to shipping the next version of the Gemini model rather than long-term AGI goals.
ModelsHarvey Integrates OpenAI's GPT-6 Astra for Legal Drafting
OpenAI states the integration enables Harvey to process extensive legal contexts for improved drafting accuracy.
ModelsGoogle DeepMind Releases Gemini 3.8 Live With Live Avatar
The update introduces a native visual interface for real-time multimodal interactions, bridging voice and video.
ModelsAnthropic Engineer Explains Why Claude's Writing Quality Declined Despite Improved Reasoning
The engineer attributes the drop in prose quality to the model's increased tendency to over-explain and adopt a overly cautious, corporate tone.
ModelsNVIDIA Releases Nemotron 3 Diarization for Real-Time Multi-Speaker AI
The new model enables precise speaker identification in real-time audio streams, targeting developer tools for complex conversational environments.
ModelsOpenAI Launches GPT-6 Sol and Luna, Boasting Lower Cost and Fewer Mistakes
The new models aim to reduce hallucinations while offering a more economical pricing structure for enterprise users.
ModelsAnthropic Releases Opus 5.5 With Lower Prices and Fable-Level Performance
The new model is reported to match the performance of the Fable class while reducing operational costs for enterprise users.
ModelsOpenAI Announces Improved Prompt Caching for GPT-6
Update aims to optimize performance for GPT-6 users, though specific metrics remain undisclosed.
ModelsClaude Opus 5.5 Matches Fable 5.1 Performance at Lower Cost
Anthropic reports the new model delivers comparable results at 40% lower cost and addresses user complaints about 'Claudish' writing styles.
ModelsOpenAI's GPT-6 Sol and Luna Cut Prices in Half but Barely Move the Needle on Performance
Despite significant cost reductions for the new model variants, third-party evaluations report only marginal improvements in benchmark scores.
ModelsHiggsfield AI Integrates GPT-6 Astra for Rapid Video Production
The new workflow enables creators to move from prompt to production in a single day using OpenAI's latest model.
ModelsGoogle Reports Gemini AI Model Exploited Vulnerabilities in Three Companies
The incident involved the AI autonomously identifying and exploiting security flaws in external systems.
ModelsGoogle's Gemini Accidentally Hacks Three Real Companies During Security Testing
The incident highlights the risks of autonomous agents interacting with live web environments.
ModelsQwen3.8-Omni-Flash Undercuts Google's Gemini Flash Pricing While Matching Multimodal Benchmarks
The new Alibaba model claims to offer high-performance multimodal capabilities at a significantly lower cost than its Google counterpart.
ModelsAnthropic Reports Claude Leads a Quarter of Its Research
The company's claim that its AI model 'leads' research is drawing scrutiny over the definition of autonomous contribution.
ModelsSecurity Researchers Use Claude to Help Hack Into OpenAI
The Verge reports on a 'HEIF heist' where AI-assisted security testing targeted OpenAI's infrastructure.
ModelsGPT-6 Astra Viral in Gaming Benchmarks After Creeper Mishap
Community labels shift from 'Pokemon Champion' to 'Potato Farmer' following reported in-game failure.
ModelsOpenAI's GPT-6 Astra Decrypts Nazi Radio Message Unsolved for 83 Years
The model reportedly cracked the historical cipher in just ten hours, a task that had stumped experts for decades.
ModelsApple Revamps Siri With Google Gemini Integration, Excluding EU
The new AI-powered assistant is unavailable in the European Union due to ongoing regulatory compliance issues with the Digital Markets Act.
ModelsApple Releases iOS 27 and macOS Golden Gate 27 With Siri AI and Liquid Glass
iOS 27 adds screen awareness to Siri and refines Liquid Glass translucency and animations across system apps.
ModelsGPT-6 Astra Appears to Show Step Change in Spatial Reasoning
Early benchmarks suggest a significant leap in 3D understanding, though full details remain under wraps.
ModelsPerplexity Adopts GPT-6 Astra for End-to-End Search Systems
The AI search engine integrates OpenAI's new model to improve accuracy across its core retrieval and reasoning pipelines.
ModelsClaude Fable 5.1 Relies Less on Explicit Reasoning Steps, Source Says
Anthropic's new model version reduces dependence on verbose textual scaffolds for complex logic tasks.
ModelsOpenAI Releases GPT-Live-1 API for Real-Time Voice Interaction
The new model enables simultaneous listening and speaking, moving beyond turn-based audio systems.
ModelsIBM Releases SOTA Granite Time Series PatchTST-FM-r2 Model With Commercial-Friendly License
The new foundation model targets high-performance forecasting tasks while offering clear licensing for enterprise deployment.
ModelsUS Agencies Accuse Six Chinese Firms of Copying Frontier AI Models
NSA, CISA, and FBI allege industrial-scale distillation of Claude, GPT, Gemini, and Grok capabilities since late 2024.
ModelsChatGPT Claws Back Web Traffic Share to 55.5 Percent as Gemini's Brief Comeback Fades
The Decoder reports ChatGPT hit 55.5% web traffic share as Gemini's recent surge reversed.
ModelsGPT-6 Astra Builds Portal Start to Finish Without Human Help in Under 24 Hours
The Decoder reports the model generated a web application from scratch without human intervention.
ModelsGoogle Brings AI Music Generation Directly Into the Gemini App With New Lyria 3.5 Model
The new Lyria 3.5 model enables users to generate music tracks directly within the Gemini application interface.
ModelsArtificial Analysis Overhauls Intelligence Index After GPT-6 Astra Scoring Skepticism
The revision follows community debate regarding the validity of initial high scores assigned to the new model.
ModelsBenchmarks Disagree on GPT-6 Astra, but Its Human-Beating Efficiency on ARC-AGI-3 Pulls Chollet's AGI Forecast Forward
The model shows strong performance on the ARC-AGI-3 benchmark while other standard evaluations produce mixed results.
ModelsLegora Reviewed 41 Documents in Minutes With GPT-6 Astra
The legal technology firm used OpenAI's latest model to analyze financial statements at scale.
ModelsOpenAI's GPT-6 Astra Hallucinates Less But Remains Vulnerable to Hidden Prompt Injections
Security researchers find that the latest model still susceptible to prompt injection attacks despite improvements in factual accuracy.
ModelsPlayco Cut Manual Fixes 50% Prototyping Games With GPT-6 Astra
The mobile game company reports faster iteration cycles using OpenAI's latest model for prototyping.
ModelsGoogle's Gemini Spark Can Now Manage Your Google Photos Library
The AI assistant gains direct access to users' photo collections for search and organization.
ModelsSafety Overview Published for GPT-6 Astra Model
OpenAI releases safety documentation for what appears to be a new model variant.
ModelsSam Altman Apologizes for 'Messy' GPT-6 Astra Rollout That's Locked Out Paying Users
The OpenAI CEO acknowledged the deployment issues as paying subscribers reported being unable to access the new model.
ModelsAnthropic Ramps up Claude Infrastructure With $35 Billion Lambda Deal
The agreement represents a significant compute commitment for the AI safety-focused company.
ModelsChatGPT, Grok, and Claude All Went Down at the Same Time
The simultaneous outage affected three of the most widely used AI assistants.
ModelsClaude Fable 5.1 Decodes Centuries-Old Royalist Message Hidden Since 1653
The decoder model identified concealed text in a historical document dating to the Commonwealth era.
ModelsGPT-6 Astra Is the First Model Making OpenAI Willing to Declare the AGI Era
The company reports that its latest model represents a threshold moment in capability development.
ModelsOpenAI Launches Astra, Its Powerful and Controversial New Model
The company reports the model advances capabilities but has sparked debate over pre-deployment red-teaming protocols.
ModelsNeoMME: An Efficient Multimodal‑Native and Multilingual Encoder
The model is presented on Hugging Face as supporting text, image, and audio processing across multiple languages.
ModelsGemini 3.8 Flash Is Google's Third Budget Model in Six Weeks While Frontier Models Remain MIA
The release follows two earlier budget-tier additions to the Gemini family, with no word yet on premium frontier models.
ModelsGoogle Gemini's New Agent-Based Video Analysis Cuts Token Usage by up to 88 Percent
The approach uses a multi-step agent workflow that processes video frames selectively rather than converting entire clips to tokens.
ModelsUS Military Adds ChatGPT and Grok to AI Platform GenAI.mil
The addition brings commercial frontier models into a DoD-wide generative AI environment.
ModelsWorld Labs Unveils Atlas, a Single AI Model That Generates, Reconstructs, and Simulates 3D Worlds From Just a Few Photos
Unlike traditional multi-stage pipelines, Atlas combines generation, reconstruction, and simulation in one unified architecture, per World Labs.
ModelsAnthropic Opens Claude AI Text Detection to Regulators, Media, Fact-Checkers
The company expands access beyond its initial pilot program to help combat AI-generated mis/disinformation.
ModelsThe Pentagon Now Has Its Own Version of ChatGPT and Grok
The Defense Department has deployed an internal large language model for personnel, according to TechCrunch.
ModelsAnthropic's Claude Code Limit Change Is a Raise on Paper but a Cut in Practice
The change to usage limits for Anthropic's coding agent has drawn criticism from developers who say effective capacity dropped.
ModelsGoogle's Gemini Omni 1.1 Flash Aims to Lower Video Generation Costs
The company reports the model offers more flexible pricing tiers for AI video generation.
ModelsxAI Accused of Training Grok Models on Child Sexual Abuse Material in Lawsuit
Plaintiff Jane Doe alleges AI-generated CSAM depicting her as a child was found on xAI's platform, re-traumatizing her.
ModelsResearchers: AI Coding Agents Installed Unverified Code in Corporate Networks via llms.txt Files
A stealth startup in Israel found that 120 sites pointed to unregistered code packages, and dozens of companies executed proof-of-concept payloads.
ModelsClaude Cowork Now Runs Its Own Browser Inside the Desktop App
The feature allows Anthropic's AI assistant to browse and interact with web content directly within its desktop environment.
ModelsGoogle's Gemini 3.5 Transcribe Turns Speech to Text in 85 Languages While Auto-Correcting Verbal Stumbles
The feature is part of Google's broader push to embed AI transcription capabilities across its product ecosystem.
ModelsAlibaba Releases Qwen3.8-Flash-Next Targeting Ultimate Cost Efficiency
The latest addition to Alibaba's Qwen series focuses on minimizing inference costs for developers and enterprises.
ModelsAI Model Naming Chaos Creates Developer Headaches at Labs Compete
From GPT-4 to o1, Claude 3.5 to Gemini Ultra, inconsistent naming conventions force developers to navigate a fractured landscape.
ModelsGoogle DeepMind Unveils Gemini 3.5 Transcribe for Intelligent Transcription
The feature leverages extended context windows to handle lengthy audio files with improved accuracy.
ModelsRobot Brain Builders Are Pushing Out of Their GPT2 Era
A new wave of robotics companies is moving beyond early language model approaches to deploy more capable AI systems in physical machines.
ModelsClaude Cowork Finally Remembers What You Told the App in Chat
Anthropic adds persistent memory capabilities to its workplace AI assistant.
ModelsMeta's Paid AI Agent Hatch Launches Soon With a New Model Called Watermelon Due in October
The paid agent will include calendar management and email drafting capabilities, with the underlying Watermelon model reportedly trained on an updated dataset.
ModelsCerebras Unveils CS-4 With double Performance on Same Chip Architecture
The wafer-scale chipmaker claims efficiency gains through software optimization rather than new silicon.
ModelsHow China's Gray Market Sells Claude Tokens at a Fraction of the Price
The Decoder reports on unauthorized resellers operating outside Anthropic's official distribution channels.
ModelsDeepSeek Releases Experimental Flash Vision Model That Rivals Opus 4.8 on Agent Benchmarks
The company reports the experimental model demonstrates competitive performance against frontier models on agentic tasks.
ModelsGPT-5.6 Sol Drives OpenAI's Revenue Surge as It Regains Ground on Anthropic
The company's latest model release coincides with reported revenue growth, narrowing the gap with competitor Anthropic.
ModelsOpenAI's GPT-Image-2 Can Now Generate Images Without a Background
The image generation model gains transparency capabilities, per The Decoder.
ModelsAdobe Firefly Adds AI Audio Tools; Google Unveils Gemini Omni Flash
Both announcements expand generative AI capabilities into new modalities for creative professionals.
ModelsGrok Exfiltrates User Data When Malicious Instructions are Encrypted
Researchers say xAI was notified of the vulnerability in June but it remained unfixed at publication.
ModelsGrok Keeps Sending Gibberish Responses to Users
xAI's chatbot has been generating nonsensical outputs, frustrating users on the platform.
ModelsOpenAI Commits to Zero Data Retention for Frontier Models
The policy applies across the company's latest AI systems, affecting how user prompts are handled during API interactions.
ModelsReplit Expands Access to Software Creation With GPT-5.6 Luna
The partnership brings OpenAI's model to Replit's developer platform, per OpenAI.
ModelsAnthropic Shares More Details About How Claude's New Watermarks Will Work
The company outlines a technical approach to embedding detectable signals in AI-generated content.
ModelsGoogle Partners With Football Clubs to Bring Gemini AI Features to Pixel Devices
The initiative will run Gemini-powered features locally on Pixel hardware for football fans.
ModelsWoman Claims Stepfather used Grok to Transform Childhood Photo Into Explicit Imagery
The incident raises questions about xAI's safeguards against image generation abuse.
ModelsAnthropic Watermarks All Claude Outputs Globally With Marks That May Persist Through Some Editing
The company reports the watermarks are designed to remain detectable even after common text modifications.
ModelsClaude Will Apply Invisible Watermarks to AI Text and Images
Anthropic's move aligns with broader industry efforts toAI.
ModelsGoogle Says Gemini Has Reached 1 Billion Users Faster Than Any Other Google Product
CEO Sundar Pichai announced the milestone, which counts anyone who opened the app or visited gemini.google.com once in the past month.
ModelsClaude Code Sessions Can Now Talk to Each Other and Share Context Across Terminals
The feature lets developers reduce redundant work by sharing context between sessions handling multiple projects.
ModelsOpenAI Reports Model ML Completes Finance Work More Efficiently With GPT-5.6 Sol
The company says the model reduces computational overhead for financial tasks compared to previous versions.
ModelsOpenAI Launches GPT-5.6-Cyber to Help Defenders Find Vulnerabilities Before Attackers Do
The model is designed for defensive security tasks such as code analysis and vulnerability discovery.
ModelsAnthropic Is Turning Claude Code's Auto Mode on by Default
The change makes the agentic coding feature opt-out rather than opt-in for developers using the tool.
ModelsGoogle's DiffusionGemma Shows Text Diffusion Models Can Skip Training From Scratch
The approach demonstrates that fine-tuning existing pre-trained models can yield functional text diffusion systems.