noticias radar ia claude android

Yesterday's radar left Google postponing its flagship model; today the game is about cost and control. Anthropic ends today — barring a third extension — the included access to Claude Fable 5, Brussels forces Google to open Android to rival assistants, and China releases the largest open model in history while Microsoft prepares an AI security audit at rock-bottom prices. Who controls the terminal and who controls the price: that is today's board.

Today is the last day of Claude Fable 5 included in your subscription (unless extended again)

The access to Claude Fable 5 included in Anthropic's paid plans (Pro, Max, Team, and Enterprise premium) expires today, Sunday, at 23:59 Pacific Time, along with the 50% increase in weekly Claude Code limits; from tomorrow, the model is paid for with separate credits at $10 per million input tokens and $50 per million output tokens. This is the third deadline — July 7, July 12, and today — and the two previous extensions were announced hours before expiration; Anthropic insists it will return the model to subscriptions when it has compute capacity. For your company: if any of your workflows use Fable 5 via subscription, decide on Plan B today — budget for pre-payment, set Opus 4.8 as a backup, or wait for the extension — and verify that no production agents change models tomorrow without your knowledge. Source

Brussels forces Google to open Android to rival AI assistants

The European Commission ordered Google on Thursday, under the Digital Markets Act, to give rival AI assistants the same access to Android system apps and services it currently reserves for Gemini, and in a second decision requires it to share search data with competing search engines; Google warns that the measures "risk the privacy and security barriers" of European users. For your company: the door is opening to choosing an assistant with deep access on Android phones — ChatGPT, Claude, or Gemini competing to be the terminal agent — and if you manage a corporate fleet, it's time to treat assistants like you treat apps: MDM policy, reviewed permissions, and an allowlist of which ones can be installed, because multiple agents with system permissions represent a new attack surface. Source

Kimi K3: China releases the largest open model in history and joins the elite

Moonshot AI launched Kimi K3, a reasoning model with 2.8 trillion parameters — the first of its kind promising open weights, before July 27 — which is already number one in the frontend coding arena ahead of Claude Fable 5, although it trailing behind Fable 5 and GPT-5.6 Sol in general indices; the API costs $3 per million input tokens and $15 per million output tokens, and in independent tests, it consumed 13,000 reasoning tokens to produce a 3,400-token response. For your company: when the weights arrive, you will be able to run a near-top-tier model on your own infrastructure or with a European provider without sending data to a Chinese API — the serious low-cost sovereignty option that was missing — but calculate the actual cost with reasoning tokens included, not just the nominal rate. Source

Microsoft prepares Project Perception: multi-model AI security to cheapen what is currently premium

According to The Information, Microsoft is finalizing Project Perception, a security product that scans code, cloud, and endpoints for vulnerabilities and proposes fixes, routing each task to the most suitable model among its own, OpenAI's, and Anthropic's to cut costs; it is expected to arrive this month as a direct competitor to Claude Mythos 5, which Anthropic sells for $10/$50 per million tokens and only to vetted partners. For your company: AI security auditing is commoditizing rapidly — if you were offered it at a premium price, wait a few weeks and compare — and the pattern Microsoft uses, routing each task to the cheapest model that solves it well, is replicable today in your own automations. Source

What to watch for tomorrow?

Tonight we will know if Anthropic extends Fable 5 for a third time — the previous two arrived hours before expiration —; the open weights for Kimi K3 (promised before the 27th), Google's formal response to Brussels, and the AI Act Digital Omnibus, which must pass through the EU Official Journal before August 2nd, remain pending.