What's new in Claude

Internal infrastructure improvements with no user-facing changes.

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

Plugins in .claude/skills directories now auto-load without a marketplace requirement; new claude plugin init scaffolding command included.

Claude Opus 4.8 is the new default model for Max, Team Premium, Enterprise pay-as-you-go, and Anthropic API accounts, with high effort by default and /effort xhigh for the hardest tasks.

Dynamic workflows let Claude write a script that orchestrates dozens to hundreds of subagents, enabling large-scale parallel agentic work from a single session.

The new security-guidance plugin reviews Claude's own changes for vulnerabilities as it works, catching issues before they land in your codebase.

Fast mode now runs on Opus 4.8 at $10/$50 per MTok, bringing the speed tier to the newest frontier model.

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

Claude Opus 4.8 launched with a 1M token context window, 128k max output, adaptive thinking, lower 1,024-token prompt cache threshold, and refusal categories in stop_details. Effort defaults to high across all surfaces.

In Claude Code: Auto mode expanded to more users, Max plan users default to fast mode on Opus 4.8, and Workflows are available as a research preview for defining multi-step agentic plans.

Mid-conversation system messages now available on Opus 4.8 — send role: "system" at non-first positions to update instructions while preserving prompt cache hits. Fast mode for Opus 4.6 deprecated (~30-day removal window).

Reset everyone's 5-hour and weekly rate limits, restoring full quotas to every Pro, Max, Team, and Enterprise subscriber. Also announced the new /goal command that keeps Claude working toward a completion condition across turns.

Highlighted Computer Use (research preview on Pro and Max), Managed Agents with Dreaming — letting agents review past sessions to self-improve memory — and enterprise admin groups with per-team capability controls.

Shared that code output per Anthropic engineer is up 200% in 2026 since launching the new Code Review feature — a team of agents that runs a deep review on every PR. Boris says he hasn't written a line of code himself this year, and ships 20–30 PRs a day running 5 parallel Claude instances.

Noted that many at Anthropic use Claude Code as a general-purpose agent, not just for code — and that he now uses it to help with almost all the work he does, not just engineering tasks.