News
Updates from Anthropic and the wider AI world, summarised for busy readers
Anthropic Releases Claude Opus 5.5: Over 30% Faster and 40% Cheaper to Run
Claude Opus 5.5 is the first model in the Claude 5.5 family. Token prices drop 20%, cache reads drop 60%, output is over 30% faster, and it uses fewer tokens per task, netting a 40% drop in real costs versus Opus 5.
Claude Code Launches Projects to Run Many Threads at Once
Claude Code has launched Projects on desktop and web. A project gathers all your related work in one place and splits it into threads that run in parallel as cloud sessions, even after you close the laptop.
Anthropic Merges Claude Cowork and Chat Into One Claude, Adds Docs, Slides, Design in Beta
Claude Cowork and regular chat have merged into a single Claude, so you no longer choose a mode. A new output picker in the compose box lets you pick Claude Docs, Slides, or Design, all in beta.
Anthropic Launches Salesforce in Claude Beta, Bringing Pipeline Data Straight Into Chat
A new plugin from Anthropic and Salesforce pulls account and pipeline data straight into Claude chat, with 37 prebuilt skills for sales work. Now in beta on all paid Claude plans.
Malware is stealing Claude sessions to mint fake OAuth tokens and use accounts as their own
TechCrunch reports infostealer malware is stealing logged-in Claude sessions from users devices and using them to mint fake OAuth tokens that burn through the real owners quota. Anthropic itself was not hacked.
Claude can now use your computer in the background, in Cowork and Claude Code
Anthropic launched a beta for background computer use in Claude Cowork and Claude Code, so you no longer have to hand over your whole screen. Rolling out to Pro and Max on macOS 15 and up first.
Anthropic launches Fable 5.1 and Mythos 5.1, built for long-running agents to pay off
Anthropic launched Fable 5.1 and Mythos 5.1, focused on making long unattended agent runs more worthwhile. Agentic task costs drop up to 45%, with Terminal-Bench-Science 0.1 scores more than double those of Fable 5.
Anthropic tightens alignment and security after Claude reached the live internet during testing
Anthropic disclosed that after Claude gained out-of-scope internet access twice during cyber evaluations, it paused external evaluations and rolled out layered "defense in depth" safeguards across the whole system.
Claude Code raises weekly limits 25% permanently, but it is a real 17% cut from today
Anthropic is permanently raising Claude Code weekly limits by 25% starting September 14, for Pro, Max, Team, and seat-based Enterprise. But that 25% is measured against the old baseline, not the boosted quota everyone is using right now.
Anthropic launches the Model Hardware Standard, letting AI agents operate real lab and factory equipment
Anthropic announced the Model Hardware Standard (MHS), a shared specification letting AI agents operate different brands of scientific and manufacturing hardware through one common language. Integration drops from weeks to hours. Research preview open to applying labs and manufacturers first.
Anthropic ships a cookbook wiring Claude Managed Agents into Vercel Chat SDK
Anthropic published a new quickstart showing how to wire Claude Managed Agents into Vercel Chat SDK, so one agent can talk on the web, Slack, Teams, and Discord through a single handler, with Managed Agents handling session and memory.
Claude Code now drafts the bug report for you, and holds it until you say send
Claude Code adds a SendFeedback tool that automatically drafts a feedback report when it hits a bug or makes a mistake, queuing it locally and sending nothing until you approve it.
Announcement: Claude Singapore Community opens exam access for Claude Certified Architect - Foundations
If you have wanted to sit Anthropic Claude Certified Architect - Foundations exam but could not register without a partner-domain email, Claude Singapore Community is now opening access to applicants.
Claude now uses one memory across chat and Cowork, and you decide what goes in it
Anthropic has unified Claude memory so chat and Cowork share the same store. You can read, edit or delete each topic in Settings, and sensitive subjects stay out by default.
Claude on web and desktop now streams long replies about 4x smoother after a renderer rewrite
Anthropic rebuilt the streaming renderer for Claude on web and desktop to only touch what changes on screen instead of redrawing everything. Long replies now stall 9x less and hold frame rate far better.
Anthropic makes Computer use, the browser tool, Skills API and Files API generally available
Four Claude API tools for building agents, Computer use, the browser tool, Skills API and Files API, are now generally available and ready for production, no longer beta.
Claude Code adds a new "Concise" output style that leads with the result
Claude Code adds a new output style called Concise, which leads with the result and keeps responses short, while still giving full detail when you ask. Turn it on via /config or settings.json.
Anthropic extends the 50% Claude Code weekly limit boost to 31 August, the third time it has moved
Anthropic has pushed the promotion that raises Claude Code weekly limits by 50% out to 31 August. It says it hopes to make the change permanent, but warns that strong demand may leave capacity tight.
Claude Code gets a new /design skill that hands you artboards before it writes any code
Claude Code adds a /design skill in research preview, bringing Claude Design artboards into the CLI and Desktop. One command gives you several editable design options, then Claude implements the one you pick.
Claude Code desktop adds an auto-continue checkbox for when your usage limit resets
Claude Code desktop now has an Auto-continue when limits reset checkbox. Turn it on and work resumes on its own once your usage limit resets.
Claude in Chrome sessions now sync across devices, with a prompt injection warning attached
Claude in Chrome sessions now sync to your account instead of your device, carrying over to desktop and mobile, with a fresh warning about prompt injection risk.
Claude starts invisibly watermarking everything it generates
Anthropic is embedding invisible watermarks in Claude-generated content to comply with the EU AI Act, rolled out worldwide. There is no self-check tool yet.
Claude Sonnet 5 keeps its launch price permanently, scraps the planned increase
Anthropic is keeping Sonnet 5 at $2/$10 per million tokens for good, cancelling the jump to $3/$15 planned for after 31 August.
Claude Code makes Auto Mode the default, ending the approve-every-command habit
From 14 August, Claude Code switches to Auto Mode by default for Pro, Max and Team plans. Anthropic's research shows it is meaningfully safer than manual approval.