Opus 4.7 ships, GPT-5.4-Cyber answers, and an unreleased model moves central banks

WEEK OF APR 11 – 17, 2026

Opus 4.7 ships, GPT-5.4-Cyber answers, and an unreleased model moves central banks

Anthropic's strongest generally-available model landed the same week OpenAI shipped a cyber-permissive variant — and the model neither of them released became the week's biggest regulatory event.

Anthropic shipped Claude Opus 4.7 as its strongest generally-available coding model and then launched Claude Design, while explicitly holding back the more cyber-capable Claude Mythos Preview.

OpenAI answered Mythos with GPT-5.4-Cyber and an expanded Trusted Access for Cyber program. Once the ECB started quizzing lenders about AI-supercharged cyber risk, the model nobody could use was the story of the week.

FRI · Apr 17, 2026fundingdeepseekchinaindustry

DeepSeek opens its first outside funding round at a $10B valuation

DeepSeek is in talks with investors to raise at least $300 million at a valuation of about $10 billion, The Information reported, citing two people familiar with the matter — its first external capital after years of self-funding by founder Liang Wenfeng's hedge fund.

Reuters could not independently verify the report, and DeepSeek did not immediately respond to a request for comment. The company has reportedly turned down earlier offers from major Chinese VC firms and tech giants. Reuters noted the round underscores the intense capital requirements of frontier reasoning and agentic models.

DeepSeek opens its first outside funding round at a $10B valuation — Anthropic

WHY IT MATTERS

The archetypal we-don't-need-your-money lab finally taking outside cash is a clean signal that reasoning models and agent systems have outgrown what even a well-funded internal backer can sustain.

Reuters The Information

FRI · Apr 17, 2026productanthropicclaudedesign

Anthropic launches Claude Design, a visual-work product powered by Opus 4.7

Anthropic Labs released Claude Design in research preview for Claude Pro, Max, Team and Enterprise subscribers. It lets users collaborate with Claude to produce designs, interactive prototypes, slides, one-pagers, wireframes and marketing collateral, applying a team's own design system by reading its codebase and design files.

Outputs can be exported, including to PPTX, or pushed into Canva via a partnership. The product is powered by Opus 4.7, Anthropic's most capable vision model at launch.

Anthropic launches Claude Design, a visual-work product powered by Opus 4.7 — Anthropic

WHY IT MATTERS

Anthropic pushing from chat and coding into visual and design surface area puts it directly against Figma-adjacent tooling — and it landed the day after the model release it depends on.

Anthropic — Claude Design

THU · Apr 16, 2026model releaseanthropicclaudeevalsvision

Anthropic ships Claude Opus 4.7 — its strongest GA model, with 3x vision resolution

Claude Opus 4.7 went generally available across Claude products, the API, Amazon Bedrock, Vertex AI and Microsoft Foundry at unchanged pricing of $5/$25 per million input/output tokens and the same 1M-token context window.

Anthropic reports a 13% lift on its internal 93-task coding benchmark, about 3x more production tasks resolved on Rakuten-SWE-Bench, and high-resolution image input up from 1,568px to 2,576px on the long edge. It adds an xhigh effort level, advisory task budgets in public beta, and a /ultrareview command in Claude Code. Anthropic is explicit that Opus 4.7 is deliberately less cyber-capable than the gated Mythos Preview, and ships with automatic blocking of high-risk cybersecurity requests.

WHY IT MATTERS

The headline model release of the week, and the first production testbed for the cyber safeguards Anthropic intends to use before any Mythos-class model ships broadly.

Anthropic — Claude Opus 4.7 CNBC

WED · Apr 15, 2026regulationcybersecurityanthropicfinance

Central banks and regulators scramble over the cyber risk from Anthropic's Mythos

Reuters reported that ECB supervisors will quiz euro-area banks on their preparedness for AI-supercharged cyberattacks from Anthropic's Mythos model, working through their regular supervisory dialogue rather than an ad-hoc meeting.

The concern follows a Treasury and Fed warning to top US bank CEOs and open warnings from UK officials, who said Mythos is substantially more capable at cyber offence than any model previously tested by the country's AI Security Institute. St. Louis Fed President Alberto Musalem said the central bank needs to revisit its cybersecurity thinking, and Barclays CEO C. S. Venkatakrishnan called Mythos a serious threat and warned there will be a Mythos 2 and a Mythos 3.

WHY IT MATTERS

A model that is deliberately not released became the week's biggest regulatory event — the clearest sign yet that frontier cyber capability, not just capability, now drives policy and supervision.

Reuters — ECB to warn bankers about Anthropic model risks Reuters — Mythos a serious threat, Barclays CEO says

TUE · Apr 14, 2026model releaseopenaicybersecuritysafety

OpenAI answers Mythos with GPT-5.4-Cyber and a broadened Trusted Access for Cyber

OpenAI unveiled GPT-5.4-Cyber, a variant of GPT-5.4 fine-tuned to be more permissive for defensive security work, and said it is scaling its Trusted Access for Cyber program from a pilot to thousands of verified individual defenders and hundreds of teams protecting critical software.

Access is tiered by identity verification: the highest tiers unlock GPT-5.4-Cyber, which lowers refusal boundaries and enables capabilities such as binary reverse engineering of compiled software without source code. Because the model is more permissive, OpenAI is starting with a limited, iterative deployment to vetted security vendors, organizations and researchers.

WHY IT MATTERS

A direct competitive and policy counter to Anthropic's gated Mythos approach — both labs are now selling cybersecurity capability through identity-gated access tiers, which is becoming the template for deploying dual-use frontier models.

OpenAI — Scaling trusted access for cyber defense Reuters

TUE · Apr 14, 2026agent toolingclaude codeanthropicdeveloper experience

Anthropic redesigns the Claude Code desktop app around parallel agents

Anthropic released a redesign of the Claude Code desktop app built to run many coding agents at once: a sidebar that manages active and recent sessions across repos, with status, project and environment filters and auto-archiving on PR merge; a drag-and-drop pane layout; an integrated terminal and in-app file editor; a rebuilt diff viewer; and in-app HTML and PDF preview.

It adds side chats for branching questions mid-task without polluting the main thread, and reaches parity with CLI plugins. The app is available to all Claude Code users on Pro, Max, Team and Enterprise plans, running sessions locally or in the cloud.

WHY IT MATTERS

Concrete tooling for the shift from working one-to-one with an agent to orchestrating a fleet of them — the same you're-in-the-orchestrator-seat framing the Opus 4.7 launch leans on.

Claude — Claude Code desktop redesign