Opus 4.8 is the new default model

Highlight

Opus 4.8 Lands as the New Default Model

Claude Opus 4.8 is now the default, with high effort on by default and a new /effort xhigh tier for your hardest tasks. Fast mode on 4.8 runs at 2x the standard rate for 2.5x the speed, a steep drop from previous Opus fast-mode pricing, and the lean system prompt is now default for all models except Haiku, Sonnet, and Opus 4.7 and earlier. Auto mode also expanded to Bedrock, Vertex, and Foundry for Opus 4.7 and 4.8 via CLAUDE_CODE_ENABLE_AUTO_MODE=1.

Worth Knowing

Dynamic Workflows Orchestrate Background Agents

You can now ask Claude to create a workflow and it will orchestrate work across tens to hundreds of background agents, with /workflows to view your runs. The post-response timer reports cumulative time when backgrounded agents or workflows are still processing, and the inline progress display was simplified to a single status row below the prompt.

Plugins Load Directly From .claude/skills

Plugins in .claude/skills directories are now auto-loaded without needing a marketplace, and claude plugin init <name> scaffolds a new one. /plugin arguments also gained autocomplete for subcommands, installed plugins, and known marketplaces, and plugins can declare defaultEnabled: false to ship disabled by default.

Run Shell Commands as Background Sessions

In claude agents, typing ! <command> runs a shell command as a background session you can attach to and detach from, also available as claude --bg --exec '<command>'. EnterWorktree can now switch between Claude-managed worktrees mid-session, and ←← to open the agents view now works on Bedrock, Vertex, Foundry, and with telemetry disabled.

Skills Gain disallowed-tools and Live Reload

Skills and slash commands can now set disallowed-tools in frontmatter to hide tools from the model while active, and /reload-skills re-scans skill directories without restarting. SessionStart hooks gained reloadSkills: true and hookSpecificOutput.sessionTitle, and a new MessageDisplay hook event can transform or hide assistant message text as it renders.

Fallback Model Now Kicks In Automatically

When your primary model isn't found, Claude Code now switches to your configured --fallback-model for the rest of the session instead of failing every request. Auto mode also no longer requires opt-in consent, and /simplify now runs a cleanup-only review focused on reuse, simplification, efficiency, and altitude rather than full bug-hunting.

Under the Hood

Credential Leak and Subagent MCP Bypasses Closed

A regression where custom API gateways could receive your Anthropic OAuth credential instead of the gateway's own token is fixed, and subagent frontmatter MCP servers now honor --strict-mcp-config, --bare, remote mode, and managed-settings allow/deny policies. Background-session subagents also no longer bypass worktree isolation to write into the shared checkout.

Long Sessions Render and Resume Faster

Long and resumed conversations are faster thanks to eliminating redundant message-rendering recomputations, terminal styling no longer degrades in very long sessions, and resuming a session by transcript path no longer eats multiple GB of memory on machines with many stored sessions.

From Anthropic

In the News