Claude's top tier ditches the 5-hour limit, renamed Ultra
Verified accounts only; the weekly 10x cap is unchanged. Also: Google folds Jules into Antigravity, and Claude reveals a hidden per-reply safety check.
Copy markdown
No more every-5-hours stall on Claude's $200 plan
Claude's $200 tier (formerly Max 20x) is now "Ultra," and verified Ultra accounts lose the 5-hour rolling session limit entirely — you run until the weekly allowance is spent instead of stalling every few hours. The catch: the weekly cap stays at 10x Pro, so total throughput is unchanged, and new accounts keep a temporary 5-hour cap until auto-verification (based on account and payment history) clears — check Settings > Usage.
Google is shutting down Jules — export your sessions now
Google is winding down Jules, its async background coding agent: it stops taking new work on Nov 11, goes read-only, and shuts off entirely on Dec 31, with its capabilities folding into the Antigravity platform. Your code and branches are safe on GitHub, but Jules-only memories, scheduled tasks, env vars, and full session history have no auto-migration and must be exported by hand before the deadline.
Antigravity opens a plugin marketplace
The same day, Google launched a Plugin Marketplace for Antigravity's app and CLI, where each plugin bundles up to five primitives — skills, MCP servers, rules, hooks, and custom agents. Launch partners include Atlassian, Figma, Stripe, Datadog, Databricks, Canva, CrowdStrike, Dropbox, and Unity, so you can wire those integrations into agent workflows instead of gluing them together yourself.
Every Claude reply reportedly gets a hidden safety self-check
Anthropic disclosed that since June's Fable 5, newer Claude models run a hidden post-response pass that reuses cached context to emit a ~10-token JSON safety verdict on each exchange. It reportedly adds under 2% inference overhead (Anthropic covers the cost) and blocks nothing in real time — verdicts feed offline aggregation and human review — while at a 1% false-positive rate it caught 78% of violations a heavier classifier flags.