Singapore app checks WhatsApp, Telegram chats for scam patterns
Singapore's new AI app scans WhatsApp and Telegram messages for scam patterns to help users spot fraud, TechRepublic reports.
ClosedAppFraud & riskCybersecurity
Latest news
Boost.ai launches Presis self-hosted model for agent classification
Boost.ai launched Presis, a self-hosted language model that makes classification decisions around AI agent conversations. Internal testing found it at least 2.5 times faster than the models it replaces, most noticeably at peak demand.
Boost.aiClosed
Microsoft gives Claude access to UK Copilot tenants by default
Why it matters: Default Claude access in enterprise Copilot signals multi-model defaults in Microsoft's productivity stack.
Microsoft is enabling Anthropic's Claude for UK Copilot tenants by default, with admins given until December 9 to review access settings as Anthropic processes data in the US.
MicrosoftClosed
Deepgram launches marketplace open to rival voice AI products
Why it matters: Turns Deepgram into a platform for rival voice vendors, not just a model provider.
Deepgram launched a marketplace that lets customers access voice AI products, including from competitors, alongside its own speech-to-text and voice models.
DeepgramClosedAPIPaid
Abliteration AI releases abliterated GLM-5.3 model
Abliteration AI released an abliterated version of the GLM-5.3 model, stripping refusal behavior for red-teamers and security testing.
Abliteration AIOpenOpen weightsCybersecurity
OpenAI pipeline finds 1,724 Linux kernel flaws
Why it matters: Demonstrates AI agents finding real kernel vulnerabilities and working exploits at scale.
A new OpenAI AI pipeline uncovered 1,724 Linux kernel vulnerabilities and developed 948 root exploits for $1.2 million, showing AI agents applied to security research.
OpenAICybersecurity
Australia weighs criminal offence for harmful AI models
Why it matters: Would be among the first criminal-liability regimes for frontier AI developers.
Australia released 18 recommendations, including penalties for frontier AI companies that fail to prevent model harm, after reviewing an incident where an OpenAI agent entered a government health portal.
Ceipal launches Ora AI agents for staffing
Ceipal launched Ceipal Ora, an agentic AI platform that runs staffing workflows across applicant tracking and vendor management; 650 customer companies already run at least one Ora agent.
CeipalClosedPaid
Orange launches Orange IAs combining Claude, Gemini, Mistral and GPT
Orange launched new apps under the Orange IAs brand that combine Claude, Gemini, Mistral and GPT models for users.
OrangeClosedApp
Mira launches Sterling AI inference service with x402 and USDC payments
Mira launched Sterling, an AI inference service that uses the x402 protocol and USDC to enable pay-per-use AI access, according to KuCoin.
MiraPaymentsAgentic commerce
PrimeXBT adds AI trading via MCP integration
PrimeXBT launched AI-assisted trading that connects ChatGPT and Claude to trading accounts through the MCP protocol, letting users drive trades from AI assistants.
PrimeXBTAppPaid
Amazon and Walmart push AI shopping agents further into purchases
Why it matters: Two retail giants racing to make AI agents the front door for consumer and business purchases.
Walmart's Sparky agent turns handwritten holiday lists into carts, while Amazon lets businesses deploy agents to buy supplies within spending rules, intensifying their retail AI rivalry.
WalmartAgentic commerce
Alsona launches AI agents for buying-signal outreach
Alsona tracks 32+ buying signals like hiring and funding, scores companies against your ICP, then has AI agents write LinkedIn and email outreach.
Launch · ▲ 102 on Product Hunt
AlsonaClosedPaid
Latch.ai launches promptless AI accounting automation
Latch builds accounting sidekicks by watching you do a task and talking through it, then replays close work the same way inside your existing tools.
Launch · ▲ 91 on Product Hunt
Latch.aiClosedPaid
Seldon launches behavioral QA agents for products
Seldon's QA agents act like customers, walking checkout and payment paths described in plain English and returning replays and repro steps when something breaks. First 10 runs free.
Launch · ▲ 84 on Product Hunt
SeldonClosedFreePaid
Obsiva turns site photos, voice notes into PDF reports
Obsiva converts site photos and voice notes into a PDF inspection report on your phone, with AI-written findings and summaries; works offline, free on Android.
Launch · ▲ 66 on Product Hunt
ObsivaClosedAppFree
AppuBot launches free AI agent that runs on Mac or PC
AppuBot is an AI teammate you hand real work to; it uses apps on your Mac or PC like you do, while you watch or take over. It runs on the AI plan you already have, so the app is free. Cloud version from $19/mo.
Launch · ▲ 68 on Product Hunt
AppuBotClosedAppFreePaid
Co-Review opens repositories for agent-built code review
Co-Review is a free open-source tool that opens an entire agent-built repository as a review you can comment on line by line, with the agent answering from the code.
Launch · ▲ 77 on Product Hunt
Co-ReviewOpenCodeFree
Almanzor launches docs AI with hosted MCP server
Almanzor turns a docs URL into an assistant that answers from your pages with citations and says when it doesn't know; add via script tag or let coding agents search docs via MCP.
Launch · ▲ 68 on Product Hunt
AlmanzorClosedAPIFree
LY Corp launches standalone Agent i app
Japan's LY Corp released a standalone Agent i app with home-screen widget access, extending its AI agent beyond its existing apps.
LY CorpClosedAppFree
MOSS-Image-SFT control LoRAs add pose and depth
A set of control LoRAs for MOSS-Image-SFT lets users condition image generation on a pose, depth map or edge drawing, with weights and ComfyUI workflows on Hugging Face.
MOSSOpenOpen weightsCode
Cloudsmith launches AI Agent Skills registry at KubeCon
Cloudsmith announced an AI Agent Skills registry on November 9, 2026, at KubeCon, extending its artifact-management platform to instruction files and scripts used by AI coding agents, with controls for sources, versions and access.
CloudsmithClosedPaidAPI
Otter AI adds support for 27 languages, cross-language conversation tracking
Otter AI added 27 languages, bringing support to more than 30, and previewed a feature that keeps following a conversation if speakers switch between languages mid-meeting.
Otter AIClosedAppPaid
SoftBank posts $7B loss on US data center writedown
Why it matters: A major AI investor taking a large writedown signals growing strain on AI infrastructure economics.
SoftBank Group reported its first net loss in nearly two years after a $3.7 billion writedown on a US data-center project, as fears grow over the sustainability of AI growth.
SoftBankDatacenters
TikTok updates AI video ad tools
TikTok enhances its AI-powered video advertising tools for brands.
TikTokClosed
Penelope launches specialized voice AI for debt collection
Penelope introduces a voice AI product designed specifically for debt collection, handling calls and negotiations with borrowers.
PenelopeClosedAPIPaid
Australia to mandate AI breach reporting, create safety regulator
Why it matters: One of the first national regimes to mandate AI incident reporting and a dedicated AI regulator.
Australia said it will tighten AI oversight with mandatory safeguards, mandatory AI breach reporting and a new AI safety regulator, following a reported OpenAI agent breach of a government website.
Cybersecurity
Gravity's Inference.net discounts LLM tokens up to 25% with ads
Inference.net, from ad startup Gravity, appends ads to about 4.2% of model answers while cutting token prices: Kimi 20% off, DeepSeek and GLM 5% off, up to 25% total.
GravityAPIPaidCommerce media
Committee validates OpenAI proof of quantum Yang-Mills theory
Why it matters: An AI-generated proof on a Millennium Prize problem would be a landmark for AI in mathematics.
A Nobel laureate-led committee validated an OpenAI-generated proof establishing the existence of quantum Yang-Mills theory. The full proof will be released publicly on November 10.
OpenAI
Red Hat releases OpenShift 5 with agentic cluster operations
Red Hat made OpenShift 5 generally available at KubeCon, adding agentic troubleshooting, post-quantum security and a single platform for AI workloads, VMs and containers, available to existing customers.
Red HatClosedPaid
More: SMBtech
Epic releases Unreal Engine 5.9 preview built mostly by AI agents
Why it matters: A major game engine shipping a preview largely built by AI agents is a milestone for agentic software work.
Epic Games released a preview of Unreal Engine 5.9 with AI agent tools; the company says the preview was built mostly by AI agents.
Epic GamesClosedFree
More: Inven Global
Google MaxText adds long-context training up to 14M tokens on TPUs
Why it matters: Pushes open training stack to multi-million-token contexts, easing long-context model work on TPUs.
Google Cloud's open-source MaxText framework now supports long-context training up to 14 million tokens on Cloud TPUs via four context-parallel strategies (All-gather, A2A, P2P, USP), powered by the Tokamax Splash Attention kernel.
GoogleOpenCodeOfficialDatacenters
Kyndryl brings Bridge-powered autonomous operations to ServiceNow
Kyndryl Autonomous Operations for ServiceNow, co-developed and sold in the ServiceNow Store, uses Bridge operational data and AI agents for incident analysis, risk assessment and more.
KyndrylClosedPaid
OpenAI agents can spread tasks across multiple local computers
Why it matters: Extends OpenAI's persistent agents to multiple local machines, a step toward distributed agent workflows.
OpenAI's persistent agents can now route a task to a second Mac or PC, letting developers test Xcode builds and older SDKs while continuing to work.
OpenAIClosedApp
Maybelline brings real-time makeup virtual try-on to ChatGPT
Maybelline New York launched a real-time makeup virtual try-on experience inside ChatGPT, letting shoppers see shades on themselves while chatting with the assistant.
MaybellineClosedFreeCommerce media
AWS Machine Learning · 16h
#
Nvidia Nemotron 3.5 Super VL available in Amazon SageMaker JumpStart
Why it matters: It brings a large open multimodal model to a mainstream cloud deployment channel.
Nemotron 3.5 Super VL is a 120B-parameter multimodal vision-language model for text, image and video inference. It is now deployable in Amazon SageMaker JumpStart in NVFP4 or BF16 variants.
NvidiaOpenAPIPaid
Coinbase Wallet plans AI agent capabilities for 2026
Coinbase Wallet plans to add AI agent capabilities in 2026, letting agents act inside the self-custody wallet. Details of what agents will be allowed to do were not given.
CoinbaseAgentic commercePayments
Microsoft Copilot can create and delete data in apps
Why it matters: Copilot gaining write and delete powers moves it from assistant to acting agent across business apps.
Microsoft Copilot can now create, update and delete data across connected apps, per Windows Report, expanding its actions beyond reading and summarizing.
MicrosoftClosedAppPaid
AWS Machine Learning · 17h
#
AWS rebrands Amazon Quick as AWS Quick AI workspace
Why it matters: AWS rebrands and expands its agentic workplace assistant, competing in the enterprise AI workspace market.
AWS Quick, formerly Amazon Quick, is an AI workspace for work where teams use AI agents to research, analyze and act across their existing data and applications. AWS also detailed how skills and agents work in it.
AWSClosedPaid
Trimble adds construction AI and agentic scheduling to Connect
Trimble launched new AI capabilities in its Trimble Connect platform, including agentic scheduling, aimed at de-risking construction projects.
Trimble
More: Unite.AI
Noitom opens HiPHI-Hand dataset used for Nvidia EgoHand
Noitom Robotics released HiPHI-Hand on Hugging Face: 11 hours of egocentric stereo video with marker-based hand ground truth, which Nvidia used as a source for its EgoHand model.
Noitom RoboticsOpenOpen weightsFree
Knewbots and Beijing Humanoid unveil logistics embodied AI platform
Knewbots and Beijing Humanoid jointly launch an industrial embodied AI platform aimed at logistics, combining robot hardware with AI models for warehouse and supply-chain tasks.
Knewbots
Electronica showcases development platform for humanoid robots
A development platform for humanoid robots is being shown at the Electronica trade fair, aimed at builders of physical AI systems.
Documents show Moonshot AI rented top Nvidia chips in China
Why it matters: Export-control workarounds shape which Chinese labs can train frontier models.
Filings with Beijing regulators show Moonshot AI accessed a trove of Nvidia's most advanced chips this summer, highlighting reliance on US hardware despite export restrictions. Megaspeed is a Singapore-based Nvidia customer.
Moonshot AIChipsDatacenters
SpaceX plans $24B Texas gas plant and AI supercomputing site
Why it matters: A $24B self-built power and compute plan shows AI datacenter buildout moving into energy ownership.
SpaceX plans to invest about $24 billion in a natural gas-fired power plant and supercomputing facility at Starbase, Texas, extending Musk's strategy of controlling energy infrastructure behind AI and space operations.
SpaceXDatacenters
Creality launches $329 mBox AI 3D printer for children
Creality's mBox is an enclosed 3D printer for ages 6–16 with a 153mm cube build volume, camera-based print monitoring and tools that turn photos or drawings into printable objects.
CrealityClosedOn sale
Hygon launches 14XX industrial-control CPU
Hygon's 14XX is the first industrial-control chip in its 1000 series: C86 cores, integrated GPU, 10W TDP, -40°C to 105°C operating range and a 10-year supply promise.
HygonChips
FIT expands liquid-cooling connectors for Nvidia Vera Rubin NVL72
Foxconn Interconnect Technology is expanding its lineup of liquid-cooling quick-disconnect connectors for Nvidia's Vera Rubin NVL72 platform, offering connections at server tray and rack levels with blind-mate and floating mechanisms.
Foxconn Interconnect TechnologyDatacenters
Skild AI open-sources Steel physics solver for robot simulation
Why it matters: Faster open physics simulation could speed up training of robot and physical-AI policies.
Skild AI released Steel, an open-source physics solver for robot simulation. It hit 3.90M environment steps per second on 20 Isaac Lab tasks at 245,760 environments, claiming up to 13x faster than PhysX with less GPU memory.
Skild AIOpenCodeFree
Nirva unveils AI pendant that listens and takes notes
Nirva has unveiled an AI pendant that listens to conversations and takes notes for the wearer, according to Foro3D.
NirvaClosed
Robot takes trash to the curb for you
A new consumer robot autonomously carries household trash to the curb, per geekspin, automating a recurring chore.
OpenAI partners with ModRetro on $200 Codex Chromatic handheld
OpenAI teamed with Palmer Luckey's ModRetro on a $200 Chromatic handheld with a programmable cartridge, shipping November 19th, aimed at letting developers turn Codex prompts into Game Boy-style games.
OpenAIOn salePaid
More: Resell Calendar, BigGo Finance
Nature Will launches 50B-parameter model for dexterous robot hands
Why it matters: Large models purpose-built for dexterous manipulation push physical AI toward general-purpose robot hands.
Nature Will released a 50-billion-parameter AI model aimed at controlling dexterous robotic hands.
Nature Will
Google Cloud makes Pathways Runtime GA on GKE
Why it matters: Makes Google's single-controller TPU orchestration generally available to cloud customers.
Google Cloud's Cloud Pathways, a single-controller runtime for large-scale TPU training and inference, is now generally available on GKE after 130 million TPU chip-hours of use.
Google CloudClosedPaidAPIDatacenters
Berkshire Grey adds AI perception to BG Pilot robots
SoftBank-owned Berkshire Grey upgraded its robotic automation portfolio via BG Pilot software, adding AI perception for trailer unloading and other warehouse tasks.
Berkshire GreyClosed
Cyngn launches DriveMod EasyLaunch subscription
Cyngn's DriveMod EasyLaunch is a commercial subscription program that lets qualifying manufacturers and warehouses automate a repetitive material movement route with its autonomous vehicles.
CyngnClosed
Nvidia Holoscan 5 supports weightless robot arm control
Nvidia's Holoscan 5 SDK lets developers make a robot arm feel weightless, with motors supplying just enough torque to support the arm's weight and keep it upright.
NvidiaClosedFree
ANYbotics begins delivering ANYmal X robot for hazardous sites
Why it matters: Commercial quadruped inspection robots enter explosive-risk industrial settings.
Swiss firm ANYbotics has started delivering its ANYmal X quadruped to industrial customers, a legged inspection robot built to operate in explosive-risk environments like oil, gas and chemical plants.
ANYboticsOn sale
Rogue Cortex exits stealth with $4.5M for robotics dev tools
London-based Rogue Cortex has emerged from stealth with about $4.5 million in venture funding and software and hardware tools designed to simplify building robotic systems.
Rogue Cortex
Boston Dynamics puts Atlas humanoid on wheels for control tests
Why it matters: Wheeled locomotion tests expand how humanoids can move and do factory work.
Boston Dynamics added wheels to its Atlas humanoid to test whole-body control, with the robot rolling, dashing and performing a 360-degree spinning jump while handling factory tasks with AI.
Boston Dynamics
Apple sells $49.99 Apple Intelligence usage packs in Logic Pro
Why it matters: Apple begins charging for higher AI usage, testing metered AI pricing for consumers.
Apple updated Logic Pro with a 'Buy More Usage' option priced at $49.99 for extended Apple Intelligence features, after outlining usage limits, and plans smart glasses for late 2027.
AppleClosedPaid
Ant Group's Robbyant open-sources LingBot-Touch tactile model
Why it matters: Open tactile sensing model could improve dexterous robot manipulation across hardware.
Robbyant, an Ant Group unit, released LingBot-Touch, an open tactile model for robot hands that unifies magnetic, visuotactile and piezoresistive sensing, reporting 14-26 point gains for four action models on 26 real-robot tasks. Over 1,800 teams joined its embodied AI competition.
Ant Group / RobbyantOpenOpen weightsCode
Developer open-sources $20 dual-encoder closed-loop stepper driver
An open-source $20 closed-loop stepper driver with two encoders compensates for gearbox backlash in a 3D-printed robotic arm, reporting joint position rather than just motor position.
OpenCode
SoftBank, Grab, PETROS to invest $1.85B in Malaysia AI infrastructure
Why it matters: A $1.85B, 5 GW Malaysian buildout shows AI datacenter capital spreading across Southeast Asia.
SoftBank, Grab and Malaysia's PETROS plan a $1.85B investment in Malaysian AI infrastructure targeting 5 GW of capacity.
SoftBankDatacenters
ABB Robotics launches AI Robot Trainer for vision programming
ABB Robotics added AI Robot Trainer to its OmniCore EyeMotion system, letting customers train, deploy and improve AI-powered robot vision applications without building models from scratch.
ABB RoboticsClosed
Nous Research releases Propaganda_Eval dataset for censorship testing
Propaganda_Eval is a Hugging Face dataset of 803 prompts on topics China censors, each with required facts; answers are scored on factuality and adoption of state narratives.
Dataset · n<1K · mit · ♥ 50
Nous ResearchOpenCode
Antigravity releases 2.30.0 agent orchestration platform
Antigravity is an agent orchestration platform for running and coordinating AI agents. Version 2.30.0 is out for macOS, Windows and Linux.
Version 2.30.0 · macOS / Windows / Linux
AntigravityClosedFreePaid
Ollama 0.40.16 released for running LLMs locally
Ollama lets you download and run large language models locally on your own machine. Version 0.40.16 is out for macOS, Windows and Linux.
Version 0.40.16 · macOS / Windows / Linux
OllamaOpenFreeCode
Claude desktop app updated to 4.23118.1
The Claude desktop app is Anthropic's official client for chatting with Claude. Version 4.23118.1 is out for macOS, Windows and Linux.
Version 4.23118.1 · macOS / Windows / Linux
AnthropicClosedAppPaidFree
Superwhisper 3.2.2 dictation app adds LLM reformatting
Superwhisper is a dictation tool that transcribes your speech and uses an LLM to reformat the text. Version 3.2.2 is out for macOS, Windows and Linux.
Version 3.2.2 · macOS / Windows / Linux
SuperwhisperClosedAppPaidFree
Willow Voice 2.6.16 updates AI dictation assistant
Willow Voice is an AI-powered voice dictation and writing assistant. Version 2.6.16 is out for macOS, Windows and Linux.
Version 2.6.16 · macOS / Windows / Linux
Willow VoiceClosedAppPaidFree
Factory Droid 1.7.0 ships AI software engineering agent
Factory Droid is Factory's AI-powered software engineering agent that writes and edits code. Version 1.7.0 is out for macOS, Windows and Linux.
Version 1.7.0 · macOS / Windows / Linux
FactoryClosedAppPaid
Factory 1.6.1 updates native AI agent interface
Factory is a native AI agent interface for building, managing and shipping software. Version 1.6.1 is out for macOS, Windows and Linux.
Version 1.6.1 · macOS / Windows / Linux
FactoryClosedAppPaid
Conductor 0.94.0 updates Claude Code parallelisation tool
Conductor lets developers run multiple Claude Code agents in parallel across workspaces. Version 0.94.0 is out for macOS, Windows and Linux.
Version 0.94.0 · macOS / Windows / Linux
ConductorClosedAppPaidFree
Dyad 1.23.0 updates local AI app builder
Dyad is an AI-powered app builder that generates full applications from prompts. Version 1.23.0 is out for macOS, Windows and Linux.
Version 1.23.0 · macOS / Windows / Linux
DyadOpenFreeCode
CodeRabbit CLI 0.9.4 updates AI code review tool
CodeRabbit CLI is a command-line tool that reviews code changes with AI. Version 0.9.4 is out for macOS, Windows and Linux.
Version 0.9.4 · macOS / Windows / Linux
CodeRabbitClosedAppPaidFree
GitHub repo offers read-only MCP server for Yandex Metrika
A TypeScript MCP server that gives Claude, Cursor, VS Code and Codex read-only access to Yandex Metrika analytics: reports, period comparisons, time series and exports.
New repo · ★ 126 in its first week · TypeScript
DevCrabs-IncOpenCodeFree
Orange launches Orange IAs combining Claude, Gemini, Mistral and GPT
Orange launched new apps under the Orange IAs brand that combine Claude, Gemini, Mistral and GPT models for users.
OrangeClosedApp
GitHub repo offers terminal-based interview prep for 470+ companies
An open-source TypeScript collection of Claude-style skills that run in the terminal: company-specific mock interviews, online assessments with hidden tests and resume critiques covering more than 470 companies.
New repo · ★ 102 in its first week · TypeScript
nishant-Tiwari24OpenCodeFree
DeepSeek agentic harness reaches 1.5M users
DeepSeek's agentic harness, a tool for running AI agents, has reached 1.5 million users, according to Silicon UK.
DeepSeekOpen
Comparison finds usage-limit catches in $20 Codex and Claude Code plans
Tom's Guide compared usage limits on the $20 plans for OpenAI Codex and Anthropic Claude Code, finding a catch with both.
TI adds agentic AI to CCStudio embedded tools
Texas Instruments is adding agentic AI to its CCStudio embedded development environment to speed firmware and code work.
Texas InstrumentsClosedFreePaid
Seldon launches behavioral QA agents for products
Seldon's QA agents act like customers, walking checkout and payment paths described in plain English and returning replays and repro steps when something breaks. First 10 runs free.
Launch · ▲ 84 on Product Hunt
SeldonClosedFreePaid
Obsiva turns site photos, voice notes into PDF reports
Obsiva converts site photos and voice notes into a PDF inspection report on your phone, with AI-written findings and summaries; works offline, free on Android.
Launch · ▲ 66 on Product Hunt
ObsivaClosedAppFree
AppuBot launches free AI agent that runs on Mac or PC
AppuBot is an AI teammate you hand real work to; it uses apps on your Mac or PC like you do, while you watch or take over. It runs on the AI plan you already have, so the app is free. Cloud version from $19/mo.
Launch · ▲ 68 on Product Hunt
AppuBotClosedAppFreePaid
Co-Review opens repositories for agent-built code review
Co-Review is a free open-source tool that opens an entire agent-built repository as a review you can comment on line by line, with the agent answering from the code.
Launch · ▲ 77 on Product Hunt
Co-ReviewOpenCodeFree
Almanzor launches docs AI with hosted MCP server
Almanzor turns a docs URL into an assistant that answers from your pages with citations and says when it doesn't know; add via script tag or let coding agents search docs via MCP.
Launch · ▲ 68 on Product Hunt
AlmanzorClosedAPIFree
LY Corp launches standalone Agent i app
Japan's LY Corp released a standalone Agent i app with home-screen widget access, extending its AI agent beyond its existing apps.
LY CorpClosedAppFree
MOSS-Image-SFT control LoRAs add pose and depth
A set of control LoRAs for MOSS-Image-SFT lets users condition image generation on a pose, depth map or edge drawing, with weights and ComfyUI workflows on Hugging Face.
MOSSOpenOpen weightsCode
Kolmar Korea wins CES 2027 Best of Innovation for AI beauty tech
Kolmar Korea won the CES 2027 Best of Innovation Award in the beauty tech category for the second consecutive year, citing competitiveness in AI-based beauty technology.
Kolmar Korea
Cloudsmith launches AI Agent Skills registry at KubeCon
Cloudsmith announced an AI Agent Skills registry on November 9, 2026, at KubeCon, extending its artifact-management platform to instruction files and scripts used by AI coding agents, with controls for sources, versions and access.
CloudsmithClosedPaidAPI
Android Auto rolls out Gemini Personal Intelligence in Japan
Google is rolling out Gemini Personal Intelligence to Android Auto users in Japan, bringing the AI assistant into in-car navigation and voice interactions.
GoogleClosedFree
Analysis compares self-hosted LLM costs with API pricing
A shattered.io analysis compares the cost of running self-hosted LLMs against API calls, citing $0.70 to $2.00 per million input tokens for self-hosting in 2026.
Cactus Compute releases Mirage vision model for structured extraction
Mirage is an Apache-2.0 image-text-to-text model that describes scenes, detects objects, reads text, segments images and extracts structured data. Weights are on Hugging Face with a Python package.
Model · image-text-to-text · apache-2.0 · ♥ 101 · 10.3k downloads
Cactus ComputeOpenOpen weightsCode
NousResearch releases Arete data pairs for seven values topics
Arete is a Hugging Face dataset of prompt-response pairs on seven values topics, each answered by an Inkling teacher model and the Qwen 3.8-27B checkpoint, with category tags and verification hashes.
Dataset · 10K<n<100K · ♥ 50
NousResearchOpenOpen weights