Today's Headlines

  • Anthropic introduces Claude Opus 5, delivering roughly Fable 5-class performance at about half the cost and becoming the new default on Max and Pro
  • Nvidia CEO Jensen Huang makes his first-ever post on X to publish a 25-signatory joint letter, "Open Weights and American AI Leadership"
  • Google signs the EU AI Act's Code of Practice on the transparency of AI-generated content

Today's edition captures two motions at once: intelligence getting a notch cheaper, and a widening argument over how that spread should be governed.

The first story is a model that brings flagship-tier performance to a lower price. The second is an industry pushing back as regulators weigh tighter rules on the open weights those models are built on. The third is a company choosing to accept a duty to label what its AI creates. The further the technology advances, the harder the question of how to govern it presses — and today it pressed on both sides of the Atlantic. We take them in order: model, policy, policy.

Today's Top Three

Anthropic introduces Claude Opus 5 at about half the cost of Fable 5

Anthropic unveiled a new model, Claude Opus 5, saying it approaches the capability of its flagship Fable 5 at roughly half the cost.

Pricing is five dollars per million input tokens and twenty-five dollars per million output tokens. Opus 5 also becomes the new default model on Anthropic's paid Max and Pro plans, bringing performance that used to sit behind higher tiers within reach of a broader set of users.

On performance, Anthropic points to a handful of figures. On ARC-AGI 3, a test of abstract reasoning, it says Opus 5 scored three times as high as the next model. On a benchmark for operating a computer, the company says Opus 5 surpassed Fable 5's best result at roughly one-third of the cost. Both are the company's own reported numbers.

Buyers can also pay for speed. For twice the base rate, a "Fast mode" responds about twice as quickly. It ships as a full feature rather than a trial beta.

Nvidia's Huang uses his first X post to publish a 25-company letter

Nvidia CEO Jensen Huang made his first-ever post on X, formerly Twitter, to publish a joint letter titled "Open Weights and American AI Leadership," signed by 25 companies and organizations.

The letter opposes a rushed tightening of rules on "open weights" — models whose weights are released publicly — and argues that keeping those weights open is what secures America's lead in AI and improves safety. That a chief executive who rarely uses the platform chose it for his opening move signals how heavily the industry weighs this fight.

The list of signatories also exposes a split within the industry. Microsoft and Meta signed the letter, while OpenAI, Anthropic, and Google did not. One camp treats open weights as a source of competitive strength; the other keeps its models closed — and the two approach regulation differently.

The backdrop is a US push to tighten rules in response to the rise of Chinese open-weight models. The administration cites allegations that Chinese firms distilled US models as grounds for regulation — a contested claim, and one this briefing has been tracking as a running story over the past week. The letter's backers counter that distillation is a legitimate technique long used across the open-source world, and that a broad ban on that basis would amount to banning nearly all open models.

Google signs the EU AI Act transparency code

Google said it will sign the EU AI Act's Code of Practice on the transparency of AI-generated content.

The code asks companies to attach a recognizable mark to AI-generated images, audio, and the like. Google says it will meet that ask by expanding its use of the SynthID watermarking technology and the C2PA provenance standard. On spreading watermarks, it plans to move in step with Apple, Eleven Labs, Kakao, Nvidia, and OpenAI.

At the same time, Google paired its cooperation with a caveat. The company warned that mounting regulatory complexity could work against the very competitiveness goals Europe has set for itself — showing, in one move, both a willingness to accept the duty of transparency and a wariness of that duty growing too heavy.

The cheaper and more widespread the technology becomes, the more closely a discipline that assigns responsibility for its output follows. In the US, industry and government are pulling on opposite ends of a rope over whether to open or close model weights; in Europe, a duty to mark generated content is coming down onto companies. Today's three stories trace both sides of that tightening around an accelerating technology.

More Moves

Model

  • Black Forest Labs, together with the robotics firm mimic, unveiled Flux 3 x Mimic, a next-generation model that combines video generation with robotic action control. By learning how the physical world behaves, it aims to handle both generation and control, and it has already been deployed at Audi for tasks such as flexible-part handling and component assembly. - Flux 3 X Mimic: The Next Generation of Video-Action Models (Black Forest Labs)

Product

  • Meta introduced a new Meta AI powered by its Muse Spark 1.1 model, positioning the assistant to plan trips, workouts, and documents and carry them from planning through app connections to completion, with a daily briefing that flags scheduling conflicts. It is rolling out in select markets first, with WhatsApp and others to follow. - Meta AI Doesn't Just Think, It Acts (Meta Newsroom)
  • OpenAI added a new voice mode to the ChatGPT desktop app. Built on its newly released ChatGPT-Live voice model, it can handle multi-step tasks — opening a thread, creating a pull request, tracking down a bug's root cause — from a single spoken command. - OpenAI's new voice mode makes it to the ChatGPT desktop app (TechCrunch)

Research

Business

  • Cognition, maker of the coding agent Devin, acquired Poke, an AI assistant used through texting apps. The price was undisclosed but described as in the "low nine figures" (between $100 million and $1 billion). Cognition wants to bring Poke's approachable, colleague-like interaction style into Devin, betting that personality — not just model quality — is becoming a real point of differentiation. - Why Cognition bought Poke: AI personality is becoming a competitive advantage (TechCrunch)
  • Midjourney acquired Co-Star, an astrology app with about 4.3 million monthly active users. Deal terms were not disclosed, and Co-Star's roughly 24-person team joins the company. It is part of Midjourney's push beyond image and video generation into new consumer categories. - Midjourney acquired the astrology app Co-Star (TechCrunch)

Source: Selected by the editorial team from the AI news inbox (collected July 25, 2026 — 29 items, 3 primary and 26 secondary).