News
Flash news aggregates global AI industry moves; columns carry original in-depth analysis from the AIBARS desk.
Moonshot AI Opens Up Its Largest AI Model Yet With Kimi K3 Public Release
Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight language model with a one-million-token context window and native vision support, for free public download, positioning it as one of the largest open AI models available to developers worldwide.
NVIDIA and SK Group Forge $500 Billion Alliance to Power Korea's AI Ambitions
NVIDIA and SK Group announced an expanded partnership worth more than $500 billion, combining a 2-gigawatt AI factory built on NVIDIA's Vera Rubin DSX platform with a long-term SK hynix memory supply and joint HBM development agreement, aiming to make South Korea a major AI infrastructure hub by 2027.
Inference Chip Startup Etched Doubles Valuation to $10.3 Billion in New Funding Round
Etched, a startup building AI chips designed specifically for inference rather than training, raised $300 million in a Sequoia-led Series C round, pushing its valuation to $10.3 billion just seven months after its last raise and following $1 billion in recent chip orders.
Jensen Huang Breaks His X Silence to Champion Open-Weight AI Models
NVIDIA CEO Jensen Huang joined X on July 24, 2026, and used his first-ever post to share an industry letter co-signed with Meta, Microsoft, Hugging Face and others, arguing open-weight AI models boost safety, innovation, and sovereignty alongside closed frontier models.
Claude Opus 5 Arrives: Anthropic's Fourth Model Launch in Two Months
Anthropic released Claude Opus 5 on July 24, 2026, claiming near-Fable 5 reasoning and coding performance at half the token price, with unchanged Opus pricing and a new low/medium/high effort control that lets developers balance cost against accuracy.
AMD Bets Billions on Anthropic in Landmark AI Chip and Equity Deal
AMD and Anthropic unveiled a strategic partnership combining up to $5 billion in AMD equity investment with a deal to deploy 2 gigawatts of Instinct MI450 GPUs, deepening collaboration on Claude optimization and ROCm software as AMD expands its role in the AI infrastructure race.
OpenAI Built an AI That Jailbreaks Its Own Models to Make Them Safer
OpenAI disclosed GPT-Red, an automated red-teaming model trained via self-play reinforcement learning that found prompt injection flaws 84% of the time versus 13% for human testers, uncovered a new 'fake chain-of-thought' attack, and helped harden GPT-5.6.
Google Skips the Flagship: Three New Gemini Flash Models Target Speed, Cost, and Cybersecurity
On July 21, 2026, Google introduced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and the security-focused Gemini 3.5 Flash Cyber, emphasizing lower token costs and agent-ready performance while confirming an active pre-training run for a future Gemini 4.
Anthropic Opens the Fable 5 Gates: Flagship Model Joins Max and Team Premium as Pro Users Pocket $100 Credits
Starting July 20, 2026, Anthropic will include Claude Fable 5 in Max and Team Premium subscriptions at up to 50% of plan limits, ending weeks of provisional access. Pro and Team Standard subscribers keep credit-based access and receive a one-time $100 Fable credit, while API pricing remains unchanged.
Moonshot AI's Kimi K3 Debuts With Trillion-Scale Parameters and Million-Token Memory
Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight mixture-of-experts model featuring a 1-million-token context window and new attention architectures, with full weights due by July 27, 2026.
Google Teams Up With Screwfix to Put AI Skills in Tradespeople's Hands
Google has partnered with UK retailer Screwfix to deliver in-store AI training pop-ups, in-depth masterclasses at Screwfix Live, and a free AI Professional Certificate for millions of tradespeople, aiming to close the gap between AI availability and everyday use in the trades.
Leaked Source Code Pulls Back the Curtain on Suno's AI Music Training Pipeline
A hacker who breached AI music company Suno shared internal source code showing how the company scraped over two million clips from YouTube Music, Deezer, Genius, and other platforms to train its generative music models.
New York Hits Pause Button on Hyperscale AI Data Center Buildout
Governor Kathy Hochul signed an executive order halting new environmental permits for data centers over 50 megawatts for up to a year, making New York the first state to impose a statewide moratorium as regulators study grid, water, and community impacts.
Anthropic Gives Claude Fable 5 Subscribers Another Week Before Credits Kick In
Anthropic extended included Claude Fable 5 access on all paid subscription tiers by seven more days, moving the cutoff to July 19, 2026, after which usage shifts to prepaid credits billed at $10/$50 per million tokens.
Apple Files Trade Secret Lawsuit Against OpenAI Over Hardware Push
Apple sued OpenAI, io Products, and two former Apple engineers, alleging a coordinated scheme spanning technical staff to leadership to obtain confidential Apple hardware and product design information.
Meta Opens Paid API With Muse Spark 1.1, Its Most Capable Agentic Model Yet
Meta launched Muse Spark 1.1 through a new paid Meta Model API, offering a 1-million-token context window and stronger agentic coding and computer-use performance, while also powering a free 'Thinking' mode in the Meta AI app.
Grok 4.5 Arrives: xAI's Cursor-Trained Model Challenges the Cost-Performance Curve
xAI released Grok 4.5, a 1.5-trillion-parameter model trained alongside Cursor, delivering competitive coding and agentic benchmark results at a fraction of the cost of comparable frontier models.
OpenAI Offers the Public a Seat at the AI Table: The $42 Billion Sovereign Fund Gambit
OpenAI CEO Sam Altman has proposed donating 5% of the company's equity — worth roughly $42.6 billion — to a US sovereign wealth fund, in a move that could reshape how the economic benefits of frontier AI are distributed to the public.
Inside Meta's $145 Billion Wager on AI Agents: Why the Payoff Is Taking Longer Than Planned
Meta raised its 2026 capital expenditure guidance to as much as $145 billion to power its push into autonomous AI agents. At a July 2, 2026 town hall, CEO Mark Zuckerberg told employees that agent development over the past four months has moved slower than expected, though he anticipates meaningful benefits within three to six months. The episode highlights the engineering challenges of agentic AI — error compounding, context degradation, and tool integration — amid a record industry-wide infrastructure buildout of up to $725 billion across the four largest tech companies.
Meta Compute Enters the Cloud Arena: Inside Meta's Bold Bid to Monetize Its AI Infrastructure
Meta announced plans on July 1, 2026 to launch Meta Compute, a new cloud infrastructure business offering GPU capacity and Llama model access to enterprises, directly challenging AWS, Azure, and Google Cloud while monetizing its $125–145B AI infrastructure investment.