Archive
Latest — page 5
Older stories, newest first.
Kimi K3 and the open-weights escalation: what 2.8 trillion parameters actually buys
Moonshot published the largest set of open weights ever released. The benchmark story is genuinely strong; the deployment story is why the frontier labs can still sleep.
US firms shift routine AI work to cheaper Chinese open models
Coinbase, DoorDash and Airbnb are routing engineering and code tasks to Moonshot, Z.ai and DeepSeek models to cut costs, Fortune reported.
OpenAI expands ChatGPT Health to all US users over 18
The rollout lets health data flow into every conversation, not just a dedicated hub, as weekly health questions hit 300 million.
Anthropic releases Claude Opus 5 at half the price of Fable 5
The new model matches near-frontier performance while cutting output costs and tightening safety guardrails.
Alphabet reports 82% cloud growth and $514 billion backlog in Q2
Google's parent company said second-quarter revenue rose 24% as AI products scaled across billions of users.
AMD pays to get into Anthropic's stack — and buys the software problem it actually needed to solve
Up to $5bn of AMD equity into Anthropic, two gigawatts of MI450 from 2027, and an engineering pact aimed squarely at ROCm. The money is the smaller half of the story.
Google develops Frozen v2 chip built around Gemini's architecture
The chip embeds parts of Gemini's model architecture directly into silicon and could handle far more tokens per unit of power than Google's newest TPUs.
Judge approves Anthropic's $1.5bn book copyright settlement
Authors and publishers will be paid roughly $3,000 per work in what is believed to be the largest copyright settlement in US history.
Microsoft builds Project Perception to rival Anthropic's Mythos in security
The platform routes tasks across Microsoft, OpenAI and Anthropic models to find and fix software vulnerabilities more cheaply.
Apple briefly overtakes Nvidia as world's most valuable company
Nvidia shares fell as much as 4.6% on Friday before recovering to close ahead of Apple by about $6 billion.
Xi launches World AI Cooperation Organization in Shanghai
Twenty-nine countries signed the founding agreement, but no major Western democracy joined the new Beijing-based body.
Thinking Machines Lab releases open-weights model Inkling
The Mira Murati-led start-up positions the 975-billion-parameter model as a base for fine-tuning rather than a top-ranked frontier system.
New York pauses permits for large data centres
Governor Kathy Hochul's executive order halts state environmental approvals for data centres of 50 megawatts or more while regulators study grid and water impacts.
Meta Puts Its Own Model Behind a Paywall: The Muse Spark 1.1 Pivot
Meta shipped Muse Spark 1.1 and, for the first time, began charging developers to use its own model via the new Meta Model API. The analyst read: a company built on free weights is now monetising inference — because openness has been commoditized out from under it.
Why Anthropic Put a Former Fed Chair on Its Benefit Trust
Ben Bernanke joins the independent body that can grow to control Anthropic's board — days after the company reportedly filed confidentially to go public. The timing is the story.
Ant's LingBot-VA 2.0 and the race to a world action model
Ant Group's robotics unit says it built the first 'embodied-native world action model' that runs on a single GPU. The claim is a simulation benchmark — and sim-to-real is where the frontier actually breaks.
AI Capital Climbs the Stack: Where July's Mega-Rounds Landed
A single week of outsized AI rounds shows investors pricing the layer above the models — agents, robotics data and deployment — while ~88% of the money stays inside the United States.
Why Google Rebuilt Gemini 3.5 Pro From Scratch Mid-Cycle
Leaked reporting says DeepMind scrapped the 2.5 Pro base and ran an entirely new pretraining pass, pushing Gemini 3.5 Pro's launch to a rumored 17 July. The analyst read: a costly, revealing bet — and a strategically ironic one.
GPT-5.6 Ships in Three Tiers, and Terra Is the Whole Strategy
OpenAI ended a 13-day government-gated preview and released GPT-5.6 as a Sol/Terra/Luna family. The frontier model is the headline; the mid-tier default and its token economics are the actual strategy.
How Anthropic Passed OpenAI on Revenue by Selling to the Enterprise
Anthropic's ~$47B annualized run-rate now runs roughly 35% ahead of OpenAI's latest disclosure. The cause is structural, not cosmetic: an enterprise-and-API business compounding faster than a consumer-subscription one.
White House races to finalise voluntary AI-release standards
The administration is negotiating pre-launch testing rules with OpenAI, Google and Anthropic. Voluntary on paper, export controls held in reserve.
Anthropic launches Claude Science, an AI workbench built for the lab
The product shipped on 30 June, but the strategy is older than that: a $400M acquisition, a Nobel hire and Amodei's 10x thesis all converge on one bet — that the moat in AI-for-science is workflow and data, not model IQ.
Meta concedes its AI restructuring underdelivered after ~8,000 cuts
A rare public admission from a company that rarely admits anything: the cuts happened, the spend is enormous, and by Meta's own account the payoff hasn't arrived. The lesson is about what headcount can and can't buy.
OpenAI proposes handing the US government a 5% stake in itself
The offer trades equity for political cover. The precedent is Intel; the ceiling is Sanders' 50%; and the whole thing is still conceptual — but the direction of travel is what matters.