metir Blog

xAI's Grok 4.7 claims a 38.0% Terminal-Bench score; Artificial Analysis measured 25.76%. Elsewhere the two sources match almost exactly. A neutral look at what the vendor says, what independent testing says, and where the two diverge.

SoftBank is raising more than $11 billion in high-yield bonds to fund a $10 billion tranche of its OpenAI investment. A neutral look at why a company borrows to buy an illiquid equity stake, and what it signals.

A Financial Times investigation found roughly $300 billion of AI infrastructure exposure held off Big Tech balance sheets through guarantees and special-purpose vehicles. How residual value guarantees work, and why they matter.

A developer teardown says OpenAI's advertising pixel and its __obi cookie can link browsing on third-party sites back to a ChatGPT account. What the mechanism is, how it compares to Meta and Google, and why the assistant angle matters.

Anthropic is merging Claude Chat and Claude Cowork into a single product that routes work on its own, and adding Docs and Slides. A neutral look at why the interface is collapsing and what it means for how people pick AI tools.

Factory raised $200 million and tripled its valuation to $5 billion in five months for its Droid coding agents. A neutral analysis of the number, the model-agnostic design, and the contested market it sits in.

Temporal raised $550 million at a $12.55 billion valuation as agentic AI moves into production. A neutral explainer on durable execution, the growth behind the number, and why long-running agents need it.

Angle Health raised $600 million at a $2.7 billion valuation to sell AI-native health plans to small businesses. A neutral look at the round structure, the profitability claim, and whether software can out-run legacy insurers.

Mozilla picked Mistral to power Firefox's Smart Window, with zero data retention and a menu of models rather than one default. A neutral analysis of the privacy posture, the open-weight bet, and why model choice is becoming the battleground.

Anthropic disclosed that Claude leads 26% of its own research and engineering work, up from under 1% in February, with about 30,000 agents running at any time. What 'leads' means, what the monitoring shows, and where the metric ends.

A zero-click remote code execution flaw named Plugin4Shell hit Claude Code, Codex, Copilot, and Gemini CLI by defeating plugin SHA pinning. How the attack works, why one design produced four vulnerabilities, and what it means for developers.

OpenAI is in early talks over a funding round that would value it above $1.2 trillion, days after Sam Altman ruled out a 2026 IPO. A neutral look at why a company this large is staying private and what that choice actually trades off.

At Dreamforce 2026, Salesforce unveiled Koa, a CRM reasoning model built with NVIDIA on an open base and post-trained on decades of CRM patterns. A neutral look at the case for narrow, domain-tuned models and where the claims need care.

AIUC raised a $40M Series A to certify and insure enterprise AI agents against a SOC 2-style standard. A neutral look at why a trust-and-insurance layer is emerging for agents, how AIUC-1 works, and what a certificate can and cannot promise.

Cohere and Germany's Aleph Alpha signed a definitive merger to build the first transatlantic sovereign AI company, valued near $20B. A neutral analysis.

Anthropic will run Claude inference from a A$31.9B Queensland data park near Dalby, taking the first of four 360 MW halls. A neutral look at the deal.

Canada and Germany pledged up to $300M to Yoshua Bengio's LawZero to build Scientist AI, a non-agentic system meant to monitor and constrain other AI.

Microsoft AI chief Mustafa Suleyman says Anthropic's approach to AI consciousness is dangerous. A neutral look at the model welfare debate and why it matters.

The Gates Foundation committed $1 billion over two years to bring AI to health, education and agriculture in poorer countries. A neutral look at the plan.

Meta is asking individual contributors in its Applied AI division to consider returning to manager roles, a partial reversal of the 2023 year of efficiency. What it says about running large AI teams.

Microsoft opened a six-week public comment window on rules for its first-party MAI models, barring cyberattacks, nuclear weapons help, and deepfakes. How a hyperscaler's self-governance compares to the labs' frontier frameworks.

OpenAI is reportedly refusing ChatGPT ads from image and audio tools that compete with its own features, surprising advertisers like Adobe. Why the category-specific carve-out matters for a one billion-user ad business.

Apple shipped Siri AI with iOS 27 on September 14, 2026, after two years of delays. What the rebuilt assistant does, and why access is gated by beta and waitlist.

Google shipped Gemini 3.8 Live and 3.8 Live Extended Thinking on Sept 15, 2026, two native speech-to-speech voice models. Here is what changed and why it matters.

Anthropic, OpenAI, SpaceXAI and Google DeepMind leaders backed slowing frontier AI. What they agreed on, what they did not, and how oversight could work.