What's new in Claude

Auto mode is now available on Bedrock, Vertex, and Foundry for Opus 4.7 and 4.8 — opt in with CLAUDE_CODE_ENABLE_AUTO_MODE=1.

Plugins in .claude/skills directories now auto-load without marketplace requirements, and claude plugin init <name> scaffolds new plugins with autocomplete support.

Fixed a bug where thinking blocks were modified when using Opus 4.8, which was causing API errors in multi-turn sessions.

Claude Opus 4.8 — Now the default model on Max, Team Premium, Enterprise pay-as-you-go, and the Anthropic API. Defaults to high effort; use /effort xhigh for the hardest tasks. Requires v2.1.154 or later.

Dynamic workflows — Claude writes an orchestration script for your task and runs it across dozens to hundreds of subagents in parallel. Use it for codebase-wide audits, large migrations, or any task too big for one conversation — managed via /workflows.

Security guidance plugin — Runs a fast pattern check on each edit, a model review at end of each turn, and a deeper agentic review on commit or push. Install with /plugin install security-guidance@claude-plugins-official.

Fast mode on Opus 4.8 — Fast mode now defaults to Opus 4.8 at $10/$50 per MTok (~2.5x faster); Opus 4.7/4.6 fast mode pricing unchanged at $30/$150. Opus 4.6 fast mode is now deprecated.

Claude Opus 4.8 launched — Most capable GA model with 1M token context window, 128k max output tokens, adaptive thinking (reasons only when needed), and a 1,024-token minimum for prompt caching. Effort defaults to high across all surfaces including the API.

Mid-conversation system messages — On Opus 4.8, you can now send role: "system" messages at non-first positions in the messages array, preserving prompt cache hits when instructions change mid-session.

Refusal categories in stop_details — Opus 4.8 now returns refusal categories when it declines a request, so applications can route different classes of refusal to the appropriate next step. No beta header required.

Claude Managed Agents on AWS — Webhooks, multiagent orchestration, and self-hosted sandboxes are now available on Claude Platform on AWS, with new IAM actions and the AnthropicSelfHostedEnvironmentAccess managed policy.

Announced dynamic workflows — Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel. Available in research preview in CLI, Desktop, and VS Code for Max, Team, and Enterprise plans, and on the Claude API, Bedrock, Vertex AI, and Foundry.

Opus 4.8 is here — the new most capable generally available Claude model. Also: Claude Design (a new Anthropic Labs product for visual outputs and prototypes) launched with Opus 4.7, and Anthropic doubled rate limits on Claude Code for developers at scale.

"Hello from Code with Claude London!" — Boris Cherny, creator of Claude Code, posted from the London leg of the Code with Claude 2026 developer conference. He's also noted he hasn't written a single line of code himself in 2026.

Quiet this week. Last notable post: responded to the HERMES.md billing bug incident in late April, announcing refunds and an additional month of credit for affected users.