Opus 5 Ships as Claude Code’s New Default, With a Security Plugin and iOS Simulator Pane
Anthropic’s latest Claude Code release makes Opus 5 the new default model, folding the faster, more cost-efficient flagship into the terminal for everyday coding work. Alongside it comes a Claude Security plugin that scans code for vulnerabilities, plus an iOS Simulator pane in Claude Code Desktop so mobile devs can build and preview without leaving the tool.
The rest of the release rounds out the agentic story: background code review, richer prompt and subagent controls, and safer auto mode and sandbox updates — landing just ahead of auto mode becoming the default on August 14. A separate bug-fix pass tackled redraw failures, Git detection on Windows, Remote Control resume leaks, self-hosted runner reliability, and memory cleanup.
Compliance API Expands to Cover Cowork and Claude Code
Anthropic extended its Compliance API to cover Cowork and Claude Code across desktop, web, mobile, and CLI, in beta for Claude Enterprise customers. Security and governance teams can now pull unified session content and metadata from the same API they already use for chat — a single pipe for audits, retention, and eDiscovery across every surface.
It’s the logical next move after yesterday’s watermarking news: as agents take more autonomous action inside Cowork and the terminal, enterprises need a way to see exactly what happened and when. Putting agentic sessions under the same compliance umbrella as chat is how Anthropic keeps regulated buyers comfortable letting Claude do real work.
Claude Sonnet 5’s Promotional Pricing Ends August 31
A heads-up for anyone with Sonnet 5 in production: the promotional rate of $2 per million input tokens and $10 per million output ends August 31. Standard pricing of $3/$15 takes effect September 1 — a 50% bump on both sides of the meter.
For high-volume workloads that’s a real line-item change worth modeling now. Sonnet 5 remains the workhorse tier between Haiku and Opus, so the move mostly hits teams running Claude at scale for classification, extraction, and agent loops. Worth a quick look at whether any traffic can shift to Haiku 4.5 without hurting quality.
Self-Hosted Claude Code Environments Reach Public Beta
Anthropic opened self-hosted environments for Claude Code in public beta, letting teams run sessions on their own infrastructure with internal network access, custom tooling, and compliance controls. It supports both fixed and on-demand runners, and is available to Team and Enterprise plans.
This is the flip side of the Compliance API story: some organizations don’t just want to audit agent sessions — they want the whole thing running inside their own walls. Self-hosted runners plus the new MCP tunnels (which reach internal servers without a public endpoint) give security-conscious shops a path to use Claude Code against private code and data with far less friction.
Claude Code’s 50% Weekly Usage Boost Extended Through August 19
Good news for heavy users: Anthropic extended the temporary 50% weekly usage boost for Claude Code subscribers through August 19, 2026. The bump gives Pro and Max coders more headroom against their weekly caps during a stretch where auto mode, background review, and Opus 5 are all pushing usage up.
It’s a small thing, but a telling one. Every new agentic feature — auto mode, background code review, subagents — burns more tokens per session. Extending the boost is Anthropic smoothing the transition so the new defaults don’t immediately run people into their limits.
Anthropic Rolls Out Claude for Teachers
Anthropic launched Claude for Teachers, entering the race with OpenAI and Google to put AI in the classroom. The product can incorporate academic standards from all 50 states, and teachers can use it to build lesson plans, personalize instructional materials, and lean on data to improve instruction. It’s free for verified U.S. educators.
Education is a deliberate wedge: high trust requirements, real workflows, and a captive audience of future power users. Pairing free educator access with the standards-aware lesson tooling is how Anthropic builds mindshare in a segment where OpenAI and Google are already spending hard.
Anthropic Is Hardening the Enterprise Stack While the Numbers Explode
Step back from today’s individual releases and one theme snaps into focus. The Compliance API now covers Cowork and Claude Code. Self-hosted environments let teams run agents on their own iron. A Security plugin scans the code Claude writes. None of these are flashy model launches — they’re the plumbing that makes a serious buyer say yes. Meanwhile the business is running hot: trackers put Anthropic’s ARR near $70B as of this month, its Series H closed at a reported $965B valuation, and a confidential S-1 is already on file.
Here’s the read: capability is table stakes now, and Anthropic is competing on deployability. Can you audit it, self-host it, secure it, and prove what it did? Opus 5 as the Claude Code default is the headline feature, but the audit trails, private runners, and security scanning are what actually convert a bank, a government agency, or a defense contractor. When your revenue is compounding this fast, the constraint isn’t model quality — it’s how quickly you can make the biggest, most cautious customers comfortable. That’s exactly where today’s releases aim.