What's new in Claude

Added enforceAvailableModels managed setting to constrain the Default model and prevent users or projects from widening managed availableModels allowlists — a new guardrail for enterprise admins.

Multiple quality-of-life fixes: mouse-wheel scroll acceleration control, /model picker improvements, Bedrock GovCloud region fixes, and performance gains for skill hot-reload and long conversations.

Fixed Fable 5 model name normalization (strips spurious [1m] suffix) and eliminated false sandbox dependency warnings on Windows.

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

Dynamic workflows let Claude orchestrate dozens to hundreds of subagents from a script it writes on the fly, enabling fully autonomous multi-step pipelines.

The security-guidance plugin reviews Claude's own changes for vulnerabilities in real time as it works — a built-in automated security reviewer for every session.

Fast mode is now powered by Opus 4.8 at $10/$50 per MTok, bringing the fastest response path up to the newest flagship model.

Claude Fable 5 (claude-fable-5) launched — Anthropic's most capable widely-released model, with a 1M token context window, 128k max output tokens, and always-on adaptive thinking. Claude Mythos 5 (claude-mythos-5) is available simultaneously for Project Glasswing participants.

Fable 5 uses the new tokenizer introduced with Opus 4.7 — the same text produces roughly 30% more tokens than pre-Opus 4.7 models. Use the token-counting API with model: "claude-fable-5" to measure prompts under the new tokenizer before migrating.

The GET /v1/environments/{id}/work endpoint for self-hosted sandboxes is now available on Claude Platform on AWS, including the new GetEnvironment IAM action that authorizes it.

Flagged requests to Claude Fable 5 now visibly fall back to Opus 4.8 — consistent with how safeguards for cyber and bio domains already behaved — making refusal boundaries more transparent.

Hosted Code with Claude: Extended Tokyo — an in-person event for indie devs and early-stage founders, featuring founder stories, builder deep-dives, and workshops from Anthropic's Applied AI team. Weekly Claude Code limits remain 50% higher through July 13 to counter Codex pricing competition.

Announced Claude Fable 5 — a Mythos-class model now safe for general use, excelling at software engineering, knowledge work, and vision. Also highlighted new Managed Agents features: "Dreaming" (scheduled memory curation between sessions) and multiagent orchestration with specialist subagents on a shared filesystem.

One-year GA retrospective: Boris Cherny says he hasn't written a line of code by hand in 8 months and some mornings coordinates tens of thousands of agents from his phone. Code produced inside Anthropic is up 8× since start of 2026. Also announced Code Review: a team of agents running deep automated review on every PR — code output per engineer up 200%, with PR reviews having been the bottleneck.

Published "We're turning Todos into Tasks in Claude Code" — a post covering the redesign of Claude Code's task-tracking system, plus recapped the Code with Claude Extended Tokyo event on June 11.