Anthropic Strikes $12 Billion AI Computing Deal

Highlight

Claude Opus 5.5 arrives as the default Opus model

Opus 5.5 (claude-opus-5-5) ships with a 1M context window at $4/$20 per Mtok and $0.20/Mtok cache reads, and it is now the default Opus model. If you are on Pro or Team Standard, your default model also moves from Sonnet to Opus, matching Max, Team Premium, and Enterprise. An effort level you saved before /effort became per-model no longer carries over to newly released models, so Opus 5.5 starts at its default until you pick one.

Worth Knowing

New command audits your prompts for older model patterns

Run /doctor prompt-audit (or /checkup prompt-audit) to check your CLAUDE.md files, skills, agents, and commands for instructions written against older models. The report leads with stale paths, stale commands, and instruction files that contradict each other.

Send now backgrounds running tools instead of cancelling

Pressing send now (Ctrl+Enter, or Ctrl+X Ctrl+S on terminals that swallow it) moves any running tools to the background rather than killing the turn. Your in-flight work keeps going while your new message goes through.

AGENTS.md and auto mode reach third-party providers

Following last week's AGENTS.md support, the fallback now also applies on Amazon Bedrock, Google Vertex AI, Microsoft Foundry, LLM gateways, and sessions with telemetry off. Those same sessions now start in auto mode when you have not configured a permission mode, and permissions.defaultMode still overrides it.

VS Code adds status, sandbox, export, and plan dialogs

The VS Code panel now answers typed /status, /sandbox, /chrome, /export, /skills, and /plan with real dialogs, including per-skill source, token estimate, and on/off state. Long replies also stop falling behind the stream now that the panel no longer re-parses the whole message on every update.

Lists gain paging, mouse support, and a prose width cap

Nearly every picker, including /tasks, /mcp, /skills, /workflows, /diff, and /plugin, now supports PgUp/PgDn, Home/End, the mouse wheel, and clicks, with scrollbars in fullscreen mode. A new maxProseWidth setting caps Claude's prose in wide terminals while tables and code blocks keep the full width.

One setting turns off all commit and PR attribution

Set "attribution": false in settings.json to drop attribution from commits and pull requests entirely. Older CLI versions skip a settings file containing the boolean form, so keep the object form in files you share across versions.

Under the Hood

Resumed sessions stop losing reasoning and prompt cache

A large batch of fixes stops --continue and --resume from re-sending earlier turns in a changed form, which had been causing the API to drop prior reasoning and break your prompt cache, and repairs histories with unreadable thinking or advisor blocks instead of failing every turn. Responses cut short by a proxy are no longer shown as complete, and duplicated stream events no longer run tool calls twice.

Startup and large-session resume get faster

Interactive startup no longer blocks on the managed-settings request, which cost about 80ms normally and 17 seconds or more on an unreachable network, and claude -p skips loading the interactive UI entirely. Resuming very large or heavily compacted sessions is also noticeably quicker, and first-request latency improves by reusing the preconnected API connection.

From Anthropic

In the News