Claude Ultra drops the 5-hour limit; Max 5x and 20x renamed
Verification criteria are unpublished and the rollout is staged, so heavy users may not see it yet. Plus GPT-6 Rosalind and a Claude red team writing exploits.
Copy markdown
No more 5-hour wall on Ultra
Anthropic renamed Max 5x → Max ($100) and Max 20x → Ultra ($200) with prices unchanged, and verified Ultra accounts lose the rolling 5-hour session window — staged from today to most existing subscribers. “Verified” is undefined, so if your usage settings still show a session timer, the change hasn’t reached your account yet.
GPT-6 Rosalind ships — if you’re let in
OpenAI added gpt-6-rosalind to the API on Oct 30, a biology-reasoning model with scientific tool use, gated to orgs approved for its Rosalind Research program. The callable GPT-6 line is the real takeaway: gpt-6.1-terra (fast visual reasoning and computer use) and gpt-6-luna behind the now-GA Decisions API, which handles up to 400 questions per request.
An unreleased Claude wrote working exploits
Anthropic’s Frontier Red Team says an unreleased research model chained real bugs into five working exploits — a zero-click root chain on Linux (CVE-2026-98432, CVSS 9.8), a one-click Chrome escape patched in Chrome 156, plus Safari, Android and Windows breaks, all disclosed before patch. The builder takeaway: frontier models can now assemble novel exploit chains with little human help, so treat agent sandboxing and dependency hygiene as load-bearing.