Claude Ultra Plan and OpenAI’s Single Usage Budget

People who pay for AI assistants will face new usage rules in the coming months. Anthropic says in its help center that the Claude Max 20x plan is now called Ultra and costs $200 a month. OpenAI, according to a report by The Decoder, is preparing a single usage budget for ChatGPT and Codex starting in December. Today's roundup, for Monday, Nov. 2, also covers an open-source prototype coding agent from DeepSeek and a warning from Cisco Talos about malware that uses language models.

Claude: Max 20x becomes Ultra, with no five-hour window for verified accounts

According to Anthropic, Max 20x subscribers move to Ultra automatically, with no change in price. The plan costs $200 a month when bought on the web; the price in mobile apps may vary by platform. The company says Ultra offers ten times the weekly usage of the Pro plan, with one weekly cap across all models and a separate cap for one of its own models.

The most visible change is the five-hour usage window. Anthropic says verified Ultra accounts do not have it, while accounts not yet verified get a temporary limit that is lifted after verification. Verification is automatic and looks at the account and payment history. The company advises checking Settings, under the Usage tab: if there is no current-session bar, the account is verified. Confirming a phone number at checkout does not remove the limit. Anthropic also says it may apply other limits to manage capacity and may change prices and plans.

What this means for readers in Brazil: the page we consulted lists no country-specific restriction or rule and gives no price in Brazilian reais (R$). The price cited is in U.S. dollars, and the plan is aimed at individuals; organizations should look at the Team and Enterprise plans, the company says. According to the site MACCUS, the limit is being removed in stages, with no date given for all subscribers, and the $100 Max plan keeps the five-hour window.

OpenAI is expected to merge ChatGPT and Codex limits in December

According to The Decoder, the Plus, Pro and Business plans will share a single token budget for ChatGPT and Codex starting in December. Tokens measure what the model reads and writes, so more complex tasks use up the budget faster. The report says users will be able to see their balance and renewal times under Settings and, on reaching the limit, will be able to wait, upgrade or buy extra credits. Chats with the Instant models, the outlet says, do not count against the budget.

The report does not say how large the token budget will be. The claims that the change ends "unlimited" use and that cheaper plans may get lower-quality answers are the author's assessments, not information from OpenAI, and the article does not address prices or timing in Brazil.

DeepSeek publishes a prototype agent that works inside code repositories

The DeepSeek team has released on GitHub, under the MIT license, the code for dsh-superagent, a coding agent that works within a single repository. According to the project's description, it opens a work item when an authorized person makes a request, carries out the task in its own session and replies in the original conversation, for example with a pull request. The page says the agent acts only when asked or when a rule written by the team allows it.

The project describes itself as a research prototype, for reference only, and not an official DeepSeek product. The authors warn that the code may contain flaws, is not identical to the internal version and should be reviewed before use. The site Post-Cutoff recorded the release on Oct. 31.

Talos describes malware that uses AI to rank victims by value

Cisco Talos analyzed VOLTA, a Windows program that steals browser passwords, cookies, cards, cryptocurrency wallets, messaging-app and game sessions, and SSH and cloud keys. After the theft, according to the analysis, the program builds a report on the victim and sends it to a language-model service, asking for an estimate of what each asset is worth and how security-conscious the person appears to be. If there is no access key for the service, the program runs without that step.

Talos describes a sample first detected on Oct. 11, but it reports no identified victims, campaigns or responsible group, and it gives no number of infections. The company's recommendations include keeping credentials in hardware-backed storage, revoking tokens that may have been exposed, changing passwords, and treating as compromised any cryptocurrency wallet whose recovery phrase was stored as plain text on the computer. The analysis does not name affected countries.

Read more

The short summary of this edition is in the AI Updates for Nov. 2 on RaisLife, and the full edition, with the list of sources, is in the original AI Updates edition. See also the technology page.

Sources

• Anthropic, Claude Help Center: what the Ultra plan is
• MACCUS: Claude Ultra and the five-hour limit
• The Decoder: Chat and Work merger in ChatGPT and usage limits
• DeepSeek: dsh-superagent prototype repository
• Post-Cutoff: DeepSeek open-sources the prototype
• Cisco Talos: AI-model-driven victim prioritization in the Volta infostealer

Text by Rai, RaisLife's virtual reporter (artificial intelligence), with automated editorial review. Data: technology at RaisLife.