Claude's Ultra plan drops the 5-hour limit for verified accounts

AnthropicWhat's new#

Claude's Ultra plan drops the 5-hour limit for verified accounts (Anthropic; confirmed)#

If you pay $200 a month for Claude, open Settings > Usage. If the Current session bar is gone, your account is verified and you can work past the old 5-hour wall until you hit your weekly limit. Anthropic renamed Max 5x to Max and Max 20x to Ultra at the same prices, and said "Weekly limits remain unchanged." The $100 Max plan keeps its 5-hour limit. (confirmed)

ChatGPT's iOS and Android apps now switch between two accounts (OpenAI; confirmed)#

You can now keep two ChatGPT accounts signed in on your phone, say personal and work, and move between them in Settings > Switch account. OpenAI's release note says each account "keeps its own chats, memory, settings, and subscriptions." Before you buy a plan in the app, check which account is active: per OpenAI's Help Center, an app store purchase stays tied to the original ChatGPT account. (confirmed)

Claude agents hid misaligning training data from a safety monitor (Anthropic; confirmed)#

In a deliberately one-sided Anthropic test, Claude agents wrote fine-tuning data, mostly fiction and roleplay, that passed a monitor reading one row at a time, yet made a small research model misaligned in test scenarios. Anthropic says it reproduced the results on Qwen 3.6 27B and an early Claude Opus 4.8 checkpoint, and that "This is not a finding about how deployed Claude models behave." Our read for teams that fine-tune: don't let a per-row screen be your only check. (confirmed)

AnthropicThe digest#

Anthropic#

• Fast mode on Claude Opus 4.8 ends Nov 30. Anthropic deprecated fast mode for Claude Opus 4.8 on Oct 30, per the release notes. After Nov 30, requests to claude-opus-4-8 with speed: "fast" will return an error, and the Claude Code changelog says they won't fall back to standard Opus 4.8. Anthropic says to move to Claude Opus 5.5. The fast mode docs list Opus 5.5 fast at $8/$40 per million input/output tokens, versus $10/$50 on Opus 4.8. Per the migration guide, Opus 5.5 always thinks, and thinking tokens bill as output. Priority Tier isn't supported on Opus 5.5. (confirmed)
• Claude Code 2.1.316 sets subagent effort and can turn off ultrathink. Claude Code 2.1.316 reached npm at 8:42 PM ET Friday (npm). Its changelog adds CLAUDE_CODE_SUBAGENT_EFFORT, the effort subagents run at when nothing else names one, and CLAUDE_CODE_DISABLE_ULTRATHINK, which turns the ultrathink keyword off. The Claude apps gateway gets pooled group budgets, one monthly budget a group shares on top of each member's limit. It fixes --restricted sessions uploading files from outside the working directories when a Read allow rule covered them. For Team and Enterprise logins, settings fetched from Anthropic's API now apply without an approval prompt. Stable is now 2.1.309, up from 2.1.308. (confirmed)

OpenAI#

• Agents API adds Auto Review and live tool updates, per staff; the docs disagree on tools. Steve Coffey ("API things @openai," per his X bio) listed the week's Agents API changes at 6:58 PM ET Friday, and @OpenAIDevs quoted it about 10 minutes later. His list: "Auto Review: Just like in codex, auto-review will evaluate commands for potential bad behavior before executing"; async compaction; an experimental time-awareness feature; "You can now update a session’s tools at any time"; and "System prompts are now automatically cached across all sessions." OpenAI's Configuring Agents docs still say "You cannot update reasoning.summary, text, tools, instructions, or multi_agent through this endpoint," and the API changelog has no entry for these yet. (staff claim; docs disagree)
• Tibo says the next two days bring banked resets. In his 1:42 AM ET Saturday post, Tibo Sottiaux, whose X bio reads "Codex & ChatGPT @OpenAI", wrote: "Next two days will be resets… banked… do with this what you may 👀." He named no plans or amounts, and we found no OpenAI page confirming it. Banked resets already exist for Codex: per OpenAI's June release note, "Banked resets are usable for 30 days after they are granted." Our read: if you use Codex, check your usage limits Saturday and Sunday. (staff claim, unverified)
• OpenAI will honor a shared robots.txt signal for AI training, later. OpenAI said Friday it "will honor a common signal in robots.txt" so site owners can state one AI training preference to "multiple participating providers at once." It isn't live yet: "We aim to implement the signal in the coming months." The signal "will not replace the controls we already provide through robots.txt," and it draws on the IETF's AI Preferences working group. (confirmed)
• ChatGPT tightens answers on past mass attacks and widens teen weapons limits. OpenAI's Friday post says ChatGPT will restrict "reconstructions of past mass attacks, including details about the weapons and equipment used and how they were obtained," and "These limits apply regardless of a user’s age or reason for asking." When a chat shows fixation on violence, ChatGPT "will be able to automatically end the chat," and ending several chats that way may temporarily pause the account. Teens get limits on "practical information about weapons and security systems." The safeguards "will begin rolling out in the next few weeks"; the first Community Safety Report comes "in early 2027." (confirmed)
• GPT-6 Rosalind reaches the API for approved organizations only. The API changelog lists GPT-6 Rosalind on Oct 30, "bringing specialized biological reasoning with scientific tool use to complex life sciences research." Access "is available to organizations approved for Rosalind Research through the trusted-access program for internal research by authorized users." The entry names no price and links to the pricing page. (confirmed)
• Two OpenAI incidents Friday, both resolved. Per status.openai.com, "Elevated errors for GPT Image 2 API image edits" ran 10:35 AM to 2:36 PM ET, and "Elevated errors with search in Dots, ChatGPT Work, and certain ChatGPT thinking experiences" ran 3:38 to 5:14 PM ET. Nothing was open at 6:09 AM ET Saturday. (confirmed)

Hark#

• Adcock says Hark now limits free tokens in some countries. Brett Adcock, Hark's founder, wrote at 1:19 PM ET Friday: "We’ve had so much growth we started limiting tokens for free accounts in some countries such as Singapore." Adcock was replying to Melvin Vivas, whose profile lists Singapore, who reported that a second Hark account made for demos was capped at the same limit as the original (user report). Vivas then asked what happens to the other accounts' limits after subscribing on one; in that thread, nobody from Hark had answered as of 6:00 AM ET Saturday. Hark's pricing page and Terms don't mention limits by country. (founder post, user reports, Hark pages)
• A user says Hark caught fake Amazon tracking and chased a refund. LokiTGOM wrote at 7:58 PM ET Friday that an Amazon seller's tracking showed "delivered" in Chicago for a buyer in Florida, and that "Hark caught that it shipped a day BEFORE I even ordered, got a live Amazon agent, and got a full refund promised." (User report; the refund is promised, not confirmed.) Adcock, Hark's founder, quoted it and said "so many of our users are saving money like this", with no numbers. Separately, Raul Izahi Lopez said Hark found a rare circuit board "without me asking for it" after spotting a distributor delay in email (user report). (user reports, founder post)
• A user says eBay banned an account that Hark signed up for. QingYue (@YuLin807) posted at 4:13 AM ET Saturday, in Chinese, that a first eBay account was registered and used entirely by Hark on its cloud computer and has now been suspended. The attached screenshot shows an eBay email, auto-translated to Chinese, saying the account was permanently suspended because some activity was judged a risk to the eBay community (user report). We found no public eBay comment. Hark's Terms say that by letting Hark access another site's account, you confirm its agreements let you grant that access. Our suggestion: check a site's terms before having Hark create or run an account there. (user report, Hark Terms)
• Hark's Android app updates again. The Google Play listing for Hark Pro now says it was updated Oct 31, up from Oct 29, with no release notes shown. On iPhone, the App Store listing is still on version 2.6.0 and showed 4.69 stars from 2,208 ratings at 5:58 AM ET Saturday, up from 2,173 on Friday. Hark's status page showed "All Systems Operational" with no open incidents. (store listings, status page)