Tuesday, May 12, 2026

Anthropic shipped Claude Opus 4.7, the latest flagship model in their lineup. The update brings meaningful improvements to software engineering and complex, long-running coding tasks, along with upgraded vision capabilities that let the model process images at significantly higher resolution. Opus 4.7 is available now across the API and Claude.ai.

Alongside Opus 4.7, Anthropic released Claude Design, a new Anthropic Labs product that lets you collaborate with Claude to produce visual outputs: designs, interactive prototypes, slides, and polished one-pagers. It's positioned as a creative co-pilot built on the new model's vision improvements.

Claude Security is now in public beta for Claude Enterprise customers. It uses Opus 4.7 to scan codebases for vulnerabilities and automatically generate proposed fixes. The launch extends Anthropic's push into agentic developer tooling beyond just code generation.

In the week's most eye-catching deal, Anthropic signed an agreement to use all compute capacity at Colossus 1, the Memphis supercluster owned by SpaceX (which now fully owns xAI). The cluster comprises over 220,000 NVIDIA H100, H200, and next-gen GB200 GPUs across more than 300 megawatts of capacity — available to Anthropic within the month. Musk told followers "no one set off my evil detector" regarding the deal.

Directly tied to the Colossus 1 deal, Anthropic doubled rate limits across Claude Code and the Claude API for Pro, Max, Team, and seat-based Enterprise tiers — effective immediately. The previous peak-hours reduction for Pro and Max subscribers has also been eliminated. A broad Claude Code update also shipped this week, adding smarter model selection, project purge tools, stronger permission handling, improved OAuth, and Windows/PowerShell fixes.

Anthropic's unreleased Mythos model — described internally as "far ahead" of any other model on cybersecurity tasks — is raising alarms among governments, banks, and utility companies. The company says it is not comfortable releasing it publicly and is limiting access to a select group of approved organizations. Senior US government officials have been privately briefed on its capabilities. The episode underscores the growing tension between frontier AI development and responsible disclosure.

Elon Musk announced on X that xAI will no longer operate as an independent company, becoming the AI division of SpaceXAI — the entity formed when SpaceX and xAI merged in a deal valued at $1.25 trillion. Grok and all xAI products now sit under the SpaceXAI umbrella, completing a consolidation that began earlier this year. The move also set the stage for the Colossus 1 compute deal with Anthropic.

The Commerce Department's Center for AI Standards and Innovation (CAISI/NIST) announced agreements with Google DeepMind, Microsoft, and SpaceXAI to evaluate frontier AI models before public release — assessing capabilities, national security risks, and cybersecurity impact. OpenAI separately said it is making its most advanced models available to all vetted levels of the US government to get ahead of AI-enabled threats.

Apple is reportedly planning a significant shift in Apple Intelligence for iOS 27, iPadOS 27, and macOS 27: users could select third-party AI providers — including Google and Anthropic — to back core system intelligence features. Integrations would flow through App Store applications. The move would open Apple's AI platform to competition for the first time and positions Anthropic's Claude as a potential default on hundreds of millions of devices.

Three converging moves this week hint at a new de-facto coding stack: OpenAI published an official Codex plugin that runs inside Claude Code; Cursor 3 (codenamed "Glass") shipped a rebuilt Agents Window for orchestrating multiple AI agents in parallel; and Codex added screen-reading and mouse/keyboard control in virtualized background sessions. A survey of 906 engineers found Claude Code is the most-loved tool at 46%, while Codex surpassed 3 million weekly active users — up from 2 million a month ago.