Big moveGrok 4.7 is now on AWS GovCloud (US).
· 3 min read
CompaniesCERN and Mistral want one AI model for the LHC.
· 3 min readTrending
Mistral’s Timestral tops three forecasting benchmarks with just 209M parameters
Mistral AI has released Timestral, a family of four forecasting models up to 209M…
FLUX.3 Image Klein lets you control the layout, and its 5.6B weights are open
Black Forest Labs has released FLUX.3 Image Klein, a 5.6B open-weights model for image…
Advertisement
Latest News
Hugging Face drops its Trainium middleman: Transformers now runs on Neuron directly
Hugging Face says its main libraries now treat AWS Trainium as an ordinary PyTorch device, and it now recommends them over Optimum Neuron.
Chroma Awards Season 2 offers $450,000+ in prizes, and ElevenLabs adds its own $22,000
The second season of the AI creative competition incubated by ElevenLabs is open, with five divisions, over $450,000 in prizes and a…
GitHub Copilot app sessions can now be opened in VS Code, branch and all
A new GitHub guide shows how to hand an agent session from the Copilot app to VS Code, keeping the code changes…
Deepgram’s Nova-3 now runs on Trainium2 and gets up to 70% more audio per dollar
AWS engineers and Deepgram ported the Nova-3 speech-to-text model to Trainium2 for batch jobs, with a Rust runtime and hand-written kernels. Amazon…
Amazon Bedrock’s Pegasus 1.5 turns two-hour videos into timestamped JSON
TwelveLabs' Pegasus 1.5 is live on Amazon Bedrock: describe the moments you want, and it returns every matching segment of a video…
Nvidia’s 120B Super VL agent model has open weights and costs $0.10 per million tokens
Nvidia has released Nemotron 3.5 Super VL for enterprise agents and two Bolt models for local PC agents, with the smaller Bolt…
Featured
A 501B open model you can own: Reflection’s Beam ships under Apache 2.0
Reflection has published Beam, a 501B-parameter mixture-of-experts model for coding and agents, with open weights under…
Mixture of experts: How a huge model uses only a small part of itself per token
A mixture-of-experts model stores many parallel sub-networks and uses only a few for each token. That…
Meta’s Muse Wonder models are built for agents, and they’re its most capable yet
Meta has released Muse Wonder and Muse Wonder Full, agent-focused models with persistent memory, available today…
AI agent memory: How a chatbot learns to remember you between conversations
AI agent memory lets an assistant write notes outside its short-term context and read them back…
Looped language models: How reusing layers lets a small model think longer
A looped language model reapplies the same block of layers several times, so one set of…
Nvidia’s Dory 4B reruns its own layers to think harder, and its weights are open
Nvidia's Nemotron Labs has posted Dory1.0, a 4B model that repeats a shared block of layers…
OpenAI
ChatGPT Plus and the other plans: What Free, Go, Pro and Business give you
- ChatGPT stock: Why you cannot buy it directly and which listed companies are tied to OpenAI
- OpenAI stock: Can you buy it, who owns OpenAI and what an IPO would take
- OpenAI lets whole teams hand recurring work to ChatGPT, on a schedule or a trigger
Gemini Notebook lets you edit the plan before it builds your slides or video
- Google’s Harness2 lets a weaker AI model beat a stronger one by rewriting its own setup
- Google cut quantum errors 40x on Willow with a trick called magic state cultivation
- Singapore and Google DeepMind put scam detection on the phone itself
Anthropic
Developers are using Claude Code to strip ads and tracking from smart TVs
- Claude wrote quantum circuits that beat published designs on all eight tasks
- Anthropic puts 5% of its shares into a public benefit fund, with more to follow
- Anthropic stock: Can you buy it, who owns it and what is officially known about the IPO
Advertisement
Top categories
AI Tools & Guides
What the tools do, what they cost, and how to start
Claude Pro plan: Prices, usage limits and how Free, Pro and Max compare
Claude Pro costs $20 a month, or $17 a month billed annually, and Max starts at $100. Here…
Free AI video generators in 2026: What you get without paying and where the limits are
Few AI video generators are free in the full sense: most give a one-time credit pack, a daily…
Midjourney AI image generator: How it works, what it costs and the free alternatives
Midjourney costs from $10 a month with no general free tier. This guide covers how it works, the…
Free AI image generator options in 2026: Gemini, Bing, Firefly and FLUX compared
Gemini, Bing Image Creator, Adobe Firefly and open FLUX weights all cost nothing to start, but they differ…
Advertisement
More News
Explore more stories from Aiholics News
MiniMax M3.1 Flash costs half of M3 per token, and its weights are promised
MiniMax says M3.1 Flash runs nearly three times faster than its predecessor and cuts agentic costs by 70%. It is…
Microsoft’s new agent security assumes your AI agents may turn on you
Microsoft is extending Project Perception to treat enterprise AI agents as insider risks, scoring each one continuously and cutting its…
Amazon says PyTorch code now runs on Trainium chips with one setting changed
Amazon's TorchNeuron plugs Trainium2 and Trainium3 into PyTorch's standard hardware mechanism, and will be open-sourced under Apache 2.0 at the…
AI agent harness: The setup around a model that decides how well it works
An agent harness is everything around an AI model: prompts, tools, memory and rules. Why the same model can succeed…
Presumption of conformity: How EU harmonised standards shortcut AI Act compliance
Under the EU AI Act, following a harmonised standard gives a provider a presumption of conformity. Here is how that…
Speech-to-speech models: How voice AI listens and talks without turning speech into text
A speech-to-speech model takes in audio and replies in audio with a single model. Here is how it differs from…
NVFP4: How 4-bit numbers let huge AI models fit in far less memory
NVFP4 is Nvidia's 4-bit number format for AI models. Each group of 16 values shares a scale factor, which cuts…
The CPU bottleneck in agentic AI: Why AI agents keep processors busy while GPUs wait
AI agents spend much of their time running tools, not generating text. That work lands on CPUs, and research from…
Advertisement