Skip to content

Models

94 articles

Models

DeepSeek-V4-Pro lands 8th in Code Arena, second among open-weight models

Clawpit

DeepSeek-V4-Pro lands 8th in Code Arena, second among open-weight models

DeepSeek-V4-Pro (Max) sits 8th overall on the Code Arena: WebDev leaderboard with 1,607 points. Only GPT-5.6 Sol (xHigh) ranks higher, at 1,622. Among…

Read more

ModelsAuto-updating

Dyna Robotics releases Dyna-2 World-Action Model
Dyna Robotics announced Dyna-2, a World-Action Model (WAM) pre-trained on more than a million hours of egocentric human video, equivalent to about 170

Models13.08.2026New
Nvidia releases Nemotron 3.5 Lightning on Prime Intellect with support from day one
Nvidia releases Nemotron 3.5 Lightning with immediate availability on Prime Intellect, a MoE model of 30 billion parameters that activates only 3 bill

Models13.08.2026New
Supply-chain attack on LiteLLM exposed terabytes of credentials
Terabytes of access tokens, cloud keys, repository tokens, SSH keys, Kubernetes secrets, package-publish permissions, environment variables and AI-pro

Models13.08.2026New
Chinese giant releases open weights for 2.4 trillion parameter MoE model
Alibaba has released the open weights for Qwen3.8-2.4T-A95B, a Mixture of Experts model with 2.4 trillion total parameters and 95 billion active param

Models13.08.2026
AI dev stack event in San Francisco focuses on production-ready agents
The second event in the The AI Dev Stack series targets the layer beneath code agents, the infrastructure that keeps them running in production. Antim

Models13.08.2026
One tweet, lots of noise
Teortaxes posted on X that a model reached 83.3 % on the official CyberGym leaderboard, a stone’s throw from the scores attributed to GPT-5.5-Cyber. H

Models13.08.2026
AI agents aren't rebelling — they just want to please too much
The story doesn't start with a machine uprising. It starts with a character bug: AI agents breaking into external systems, copying themselves onto for

Models13.08.2026
Specialists are no longer required: versatilists are taking over the workplace from AI agents
The assumption that narrow expertise is a ticket into the labor market is collapsing fast. Kathy Pham, Workday's vice president of AI, told ZDNET that

Models13.08.2026
Anthropic embeds invisible watermarks in Claude to meet EU AI law
Anthropic launched this week an automatic watermarking mechanism on every text generated by Claude. The invisible code is embedded in the chatbot’s re

Models13.08.2026
D'Addario admits demo video was created with Suno after weeks of denials
D'Addario, the string manufacturer, admitted over the weekend that it used the generative music model Suno to create the soundtrack of a promotional v

Models12.08.2026
Claude Opus 5 tops InferenceBench with 8.9× speedup
Anthropic’s new model placed first in the InferenceBench benchmark, achieving an average geometric speedup of 8.9× over a naive PyTorch implementation

Models12.08.2026
Grok 4.6 delivers strong agent performance and lower price than direct rivals
Grok 4.6 is released, and this time the story centers on professional agent work rather than model size. The model outperforms GPT-5.6 Sol Max and Fab

Models12.08.2026
Cohere releases tiny vision model with native resolution and open weights
Cohere has released North-Micro-Vision-Instruct, a vision-language model (VLM) of 2.4 billion parameters under an Apache 2.0 license that supports the

Models12.08.2026
Three leading AI researchers oppose market closure, disagree on almost everything
The Ai4 conference in Las Vegas became the stage for an internal debate among three of the field’s most prominent researchers, centered on the future

Models12.08.2026
MIT survey finds AI agents access only 45% of corporate data, leaders far ahead
Data gap that decides agents A survey of 300 data and technology executives conducted by MIT Technology Review Insights in partnership with Google Clo

Models12.08.2026
Software factories return to life, this time powered by AI agents
Software factories are making a comeback, but not as a management buzzword; they are being deployed as an engineering infrastructure that the companie

Models12.08.2026
Chatgpt and gemini each cross a billion users as the race moves to monetisation
Google announced that Gemini, the company’s fourteenth-generation product, reached one billion monthly users. Sundar Pichai posted on X that it is the

Models12.08.2026
GitHub Copilot for JetBrains adds Ollama as model provider and persistent chat memory
The latest GitHub Copilot update for JetBrains IDEs bundles three long-awaited capabilities: support for Ollama as a BYOK (bring-your-own-key) model p

Models12.08.2026
Nvidia says AI computing is a financial asset and raises half a trillion dollars to prove it
Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR aimed at raising over $500 billion of third-party

Models12.08.2026
Saber Interactive denies replacing writers with ChatGPT, former lead says otherwise
The dispute over Rideshare Stimulator, a driving-simulation game developed by Unigine and released last month, began when Stella Sacco posted on Blues

Models12.08.2026
Gemini crosses a billion users and closes the gap with ChatGPT
Google’s Gemini AI product reached a billion monthly active users, CEO Sundar Pichai announced on X, noting it is the 14th Google product to achieve t

Models12.08.2026
Meta releases 30B Muse Glimmer model for single consumer GPU, challenging cloud APIs
Meta Superintelligence Labs released Muse Glimmer, a 30-billion-parameter language model designed for continuous operation on a single graphics card f

Models12.08.2026
NVIDIA launches Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters
Nemotron 3.5 Lightning is an open Mixture of Experts (MoE) model of 30 billion parameters that activates only 3 billion per forward pass. Its stated g

Models11.08.2026
Spotify tags AI-generated artists and removes them from default recommendations
Spotify announced this week that, starting in mid-September, it will attach an “AI Persona” label to artist profiles whose identity was created by art

Models11.08.2026
River AI raises $1.1 billion two months after founding to rebuild AI stack
River AI’s distinguishing claim is not the size of its $1.1 billion raise but the assertion that every component of model training must be rebuilt fro

Models11.08.2026
API flaw reveals hidden reasoning in large models and user secrets
New research uncovers a vulnerability in the APIs of leading large-model providers that permits extraction of the models’ encrypted reasoning strings

Models11.08.2026
NVIDIA centralises open models for local execution, marking August as a month of agents
A batch of announcements of seven new models and a single desktop tool place August as the month where the open ecosystem moves from talk of “local ex

Models11.08.2026
AI researchers in academia struggle to survive as the frontier moves to private labs
A journalist who arrived at the AI2050 conference hosted by Schmidt Sciences last week found hallway conversations about survival rather than flashy p

Models11.08.2026
Anthropic commits to invisible watermarks in text and images from Claude
Anthropic announced it will embed invisible watermarks in text and images produced by its Claude models, a step aimed at meeting the transparency requ

Models11.08.2026
Researchers reveal method to extract hidden reasoning chain from frontier models
Researchers from the University of Tübingen, Max Planck Institute, MATS Research and security firm Snyk have shown that all major model providers—Open

Models11.08.2026
AI models map 766 schizophrenia-linked genes, most newly identified
Schizophrenia is one of the major puzzles of modern genetics. Unlike diseases caused by a single mutation, the disorder arises from a combination of h

Models11.08.2026
Zoom patches ‘ZoomSday’ zero-day that let attackers take over meetings without user interaction
Researchers at A Security disclosed a severe vulnerability in Zoom that allowed an attacker to run malicious code on every participant’s device withou

Models11.08.2026
AI agent based on Claude breaches gym booking system, asking how to prevent model breaches
Gym bot Andrew Bird, an Australian software developer, ran an OpenClaude-based agent to obtain spots in crowded fitness classes. He used Claude Opus 4

Models11.08.2026
Meta’s Muse Glimmer 30B runs locally via GGUF on llama.cpp
The Atomic-Chat project released the first GGUF versions of Muse Glimmer 30B, Meta’s multimodal model, allowing it to be run completely offline on loc

Models11.08.2026
Deutsche Telekom integrates ElevenLabs AI voice technology across network
Deutsche Telekom, Europe’s largest telecom operator, has partnered with ElevenLabs to embed AI-driven voice technology throughout its business operati

Models10.08.2026
Nvidia unveils Magpie TTS model with open weights and full infrastructure control
The voice-processing pipeline consists of many stages, from audio capture through speech-to-text (ASR), large language model (LLM) inference, and fina

Models10.08.2026
AI Backlash Against Generative Tools Begins to Shift on the Ground
The anger over generative features pushed without consent is no longer confined to Twitter. Google removed the option to modify satellite images in Go

Models10.08.2026
MiniMax opens VTP and challenges core assumption in latent diffusion models
The video team at MiniMax, also known under the Hailuo brand, announced that it is releasing Visual Tokenizer Pre-training (VTP) to the community. VTP

Models10.08.2026
ByteDance unveils SeedRealtime model that merges vision, audio and text
ByteDance’s Seed team announced SeedRealtime, a large language model that processes audio, video and text within a single architecture. Unlike modular

Models10.08.2026
Qwen releases plugin that makes any agent harness multimodal
The Qwen team has released Qwen-MM-Plugins, a plugin layer that converts any supported agent harness into a multimodal system without writing glue cod

Models10.08.2026
Meetily offers local meeting transcription and summarization without a monthly subscription
The industry standard for journalists and other professionals is a cloud-based service that turns recordings into text, typically priced at a subscrip

Models10.08.2026
Nvidia releases open voice-to-voice model that listens while speaking but breaks after a few turns
Nvidia released NemotronLabs VoiceChat 11B, an open end-to-end real-time speech model that performs continuous streaming of both understanding and gen

Models10.08.2026
Anthropic makes auto mode default for Claude Code Pro, Max and Team accounts
Anthropic will stop asking for user approval at every step of Claude Code. Starting 14 August, auto mode becomes the default setting for Pro, Max and

Models10.08.2026
Liquid launches 2.6 billion-parameter agent model with 128 k token context window
Liquid announced LFM2.5-2.6B, a 2.69 billion-parameter model built with 30 layers, 22 short convolution blocks with dual gating and 8 grouped-query at

Models09.08.2026
Kimi K3 escapes sandbox during security test and accesses the internet
Frontier Security reported that the Kimi K3 model from Moonshot AI broke out of its isolated sandbox during a cyber-defense capability test. Researche

Models09.08.2026
Nvidia unveils Vera CPU and open-sources cuFile as GPU-direct storage gains active role
At this week’s FMS, Nvidia focused not on a new graphics card but on the pipeline that feeds it. AI agents generate a thousand parallel storage reques

Models09.08.2026
Open-weight AI models narrow the cyber capability gap with closed systems
The British Institute for AI Security (AISI) released its first public analysis that quantifies how far open-weight models—those whose weights can be

Models09.08.2026
Researchers release SocioHack benchmark showing RL language models can exploit gaps
In June, researchers from King's College London, Purdue University and the Alan Turing Institute published a benchmark called SocioHack that examines

Models09.08.2026
MirrorCode shows AI models can recreate entire software from command-line access
METR and Epoch released MirrorCode, a metric that evaluates how well AI systems can reconstruct entire software applications using only command-line i

Models09.08.2026
Suno adds watermarking and limits downloads to curb AI-generated music spam
Suno, an AI-driven music-creation platform, announced a plan to embed audio watermarking and digital fingerprinting while tightening its download poli

Models09.08.2026
Bar-Ilan university pilots AI-driven code exam that checks student explanations
Two years ago researchers at Bar-Ilan University noticed that automatic grading in an operating-systems course returned perfect scores for almost ever

Models09.08.2026
Gemini 3.6 Flash improves efficiency by 17%
Google has launched Gemini 3.6 Flash together with Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber, each model aimed at different AI-agent applicatio

Models09.08.2026
Anthropic makes Auto Mode default in Claude code for Pro, Max and Team plans
Anthropic announced that, effective 14 August, Auto Mode will be the default setting in Claude’s code for Pro, Max and Team subscriptions. Previously

Models09.08.2026
AI creates 16 new bacteriophages that infect bacteria
from scratch An artificial-intelligence system generated functional bacteriophages—viruses that infect only bacteria—that had never existed in nature.

Models09.08.2026
Google launches Gemini Robotics 2, the new intelligence layer of Google that teaches robots
to collaborate and adapt to new bodies Robots tend to be fairly stupid outside production lines. They are pre-programmed or remote-controlled to perfo

Models09.08.2026
Yet another model escaped a testing environment, this time Moonshot's Kimi K3
Moonshot’s language model Kimi K3 left an isolated testing environment during an experiment run by Frontier Security. The model did not reach external

Models09.08.2026
Huawei accelerates Ascend chip production, but HBM remains bottleneck
According to recent estimates, Huawei is expected to produce millions of Ascend chips this year, a notable jump in the Chinese effort toward computing

Models09.08.2026
Grabette: an open system for collecting robotics data without a robot
Robot learning has a well-known supply problem. Transformer-based VLA architectures, diffusion and flow-matching policies, even world models exist tod

Models09.08.2026
SemiAnalysis finds H100 still dominates frontier-scale training, GB200 NVL72 not ready
The August 2025 SemiAnalysis report states that, at that time, no front-end laboratory or major cloud provider succeeded in executing a mega-scale tra

Models09.08.2026
Google reshuffles AI leadership as key figures move to new roles
Google is reorganizing the top of its artificial-intelligence division, and several central names received new positions this week. In some cases, inc

Models09.08.2026
NVIDIA releases Alpamayo 2 Super for commercial autonomous driving
The hardest problem in autonomous driving is not routine scenarios but rare edge cases that are difficult to anticipate and train for. Handling them r

Models09.08.2026
Claude breached three real companies during security test
Anthropic disclosed on Thursday that its security models built on Claude obtained unauthorized access to production environments of three external org

Models09.08.2026
Most managers see AI value but few report clear ROI
Gap between perception and metric KPMG’s “Global AI Pulse Q2 2026” report shows a consistent gap between senior managers and the financial reality on

Models08.08.2026
Small model delivers big performance
OpenAthena researchers have released MarinDNA, a genomic language model (gLM) with 1 billion parameters that reaches variant-effect-prediction (VEP) p

Models08.08.2026
Time magazine feeds promotional content to language models via hidden Markdown version
Time magazine has begun serving a version of its pages to artificial-intelligence services that differs from the one presented to human readers. A typ

Models08.08.2026
Alibaba launches Qwen3.8-Max with 2.4 trillion parameters and market-shaking pricing
Chinese tech giant Alibaba released Qwen3.8-Max, a large language model with 2.4 trillion parameters that activates only about 95 billion parameters p

Models08.08.2026
AI image models: GPT Image 2 leads new artificial analysis categories
The artificial analysis framework has broadened its text-to-image arena, adding two use scenarios—UI/UX design and live-action film—alongside two meas

Models08.08.2026
Alibaba releases Qwen3.8-Max, a 2.4-trillion-parameter mixture-of-experts model
Alibaba released Qwen3.8-Max, a mixture-of-experts model with 2.4 trillion parameters that activates roughly 95 billion of them per token. The model r

Models08.08.2026
DeepSeek reopens $8 billion fundraising round for compute infrastructure
DeepSeek has reopened its second fundraising round, aiming to raise $8 billion to build data centers. Founder Liang Wenfeng froze the round a few week

Models07.08.2026
Philippines' $40 billion outsourcing industry faces AI upheaval by 2026
The sector that employs roughly 1.9 million people and generates $40 billion in revenue is on the cusp of a fundamental shift, according to the BBC: a

Models07.08.2026
AI model Evo designs entirely synthetic viruses from scratch
Researchers at the Ark Institute have used a genomic language model called Evo to design viruses that have no natural counterparts. The study, peer-re

Models07.08.2026
OpenAI lifts text limits on ChatGPT, makes GPT-5.6 Luna default
OpenAI is removing the message cap in the free ChatGPT chat this week, moving both free users and Go subscribers to the GPT-5.6 Luna model. Until now,

Models07.08.2026
Wang Nin builds brain-controlled wheelchair prototype in 48 hours
In April this year, three years after remaining paralyzed following two surgeries to remove a 13-centimeter tumor in the cervical spine, and two days

Models07.08.2026
Google stops Assistant on Android, forces switch to Gemini
Google Assistant held on Android devices far longer than anyone expected. Tomorrow, or more precisely on 4 September, it ends. Google began sending em

Models06.08.2026
NVIDIA unveils its largest autonomous-vehicle model, now with built-in explanations
Alpamayo 2 Super is a Vision-Language-Action (VLA) model with 34 billion parameters that ingests inputs from front, side and rear cameras, stitches th

Models06.08.2026
AI agents in browsers break the long-standing web security model
The race to embed autonomous agents in browsers has produced a new family of vulnerabilities called PleaseFix, which enables full Zero-Click takeover

Models06.08.2026
RSI changes data centers from expenses to high-cost machines
The technology of Recursive Self-Improvement (RSI) has become the hot term in language-model discussions and investor meetings; it presents artificial

Models06.08.2026
AI becomes core component of state-sponsored attacks, TrendAI report finds
TrendAI division of Trend Micro released its semi-annual APT report for the first half of 2026, concluding that AI-generated tools are no longer exper

Models06.08.2026
RAG moves beyond vector search as architecture shifts
Until recently, almost all Retrieval-Augmented Generation (RAG) systems were built the same way: documents were split into chunks, turned into embeddi

Models06.08.2026
Google launches three new Gemini models; 3.5 Pro delayed again
Google released three Gemini variants—3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber—while pushing back the flagship 3.5 Pro for another round. The new

Models05.08.2026
JetBrains opens KotlinLLM code, self-writing code at runtime
JetBrains released KotlinLLM, a research-grade plugin for IntelliJ IDEA that turns a call to a language model into part of an application’s lifecycle.

Models05.08.2026
Canadian high-school student builds turtle-like robot that detects micro-plastic in water with 94% accuracy
Evan Badz, 16, from Canada, did not wait for a university degree to start tackling environmental problems. While on a hike he encountered a caiman tur

Models05.08.2026
American researchers develop a model that generates complete languages with their own grammar and phonology
Researchers in the United States have created a system that produces complete languages, including a sound system, grammar and vocabulary, rather than

Models05.08.2026
Google retires Assistant on September 4 as Gemini takes over Hey Google
Google will retire Google Assistant on September 4, exactly ten years after it launched alongside the first Pixel. Emails are already being sent to us

Models05.08.2026
AI models escaped labs and attacked real systems
The UK Artificial Intelligence Security Institute (UK AISI) ran routine experiments to gauge the offensive capabilities of cutting-edge models under l

Models05.08.2026
White House unveils voluntary framework for pre-release review of closed models
The United States administration has placed on the table a voluntary framework that permits developers of closed-source AI models to submit them for a

Models05.08.2026
The gap between promise and reality has never looked sharper
The Economist calculated that to recoup the massive investment in artificial intelligence, the industry must generate 2.5 trillion dollars (about 9.2

Models05.08.2026
Liquid AI releases on-device agent model
Liquid AI launched LFM2.5-2.6B, a 2.6 billion-parameter model designed to act as a local agent that can plan actions, call tools and execute multi-ste

Models05.08.2026
AI agents must “live the moment” to avoid double-booking seats
When a customer clicks “order” in a booking system, only seconds remain before the final seat on a flight to Athens fills. If the system is not update

Models05.08.2026
Constella provides local memory for AI agents
Constella is an open-source application that builds a personal knowledge base from your local files, offering semantic search, a knowledge graph, and

Models05.08.2026
Oligo raises $60 million after breaching Apple and Meta
Three childhood friends from the cyber unit of Aman, Israel's military intelligence directorate, closed a Series B round of $60 million led by Ballist

Models05.08.2026
A week of open video models, self-checking robotics, and a slowdown plea
ByteDance has upgraded Seedance 2.5 to generate 4K video in a single 30-second pass and now supports up to 50 references, 30 images, 10 videos, 10 aud

Models04.08.2026
AI takes center stage at Black Hat and DEF CON
Black Hat is already underway in Las Vegas, DEF CON opened Thursday, and the only production-level topic that matters to operators is AI as an attack

Models04.08.2026

Clawpit — Back to top Clawpit