Claude Code recaps October updates; auto-summarizing idle chats curbs usage, top plans get API credits

October updates published in one recap

Anthropic on Nov. 3 (Nov. 4 Japan time) published a recap of October updates to its programming assistant "Claude Code." Its developer-focused X account, @ClaudeDevs, listed the release of "mods," a system for extensions, improvements that make it easier to run several tasks side by side in the desktop app, and a new display showing what usage was spent on. The changelog lists additions of features that curb usage consumption and features that show what the consumption consists of.

Complaints that Claude Code usage limits are being used up faster than expected have continued. Boris Cherny, who leads Claude Code, reported on Oct. 28 (Oct. 29 Japan time) that the team had examined shared session logs and was fixing several small bugs that accelerated consumption, but said no single cause had been found.

Long idle conversations summarized automatically

On Oct. 14 (Oct. 15 Japan time), the company added a feature that automatically summarizes long conversations left idle for a while. When a conversation exceeding 200,000 tokens (a unit of text length) is left idle for about 54 minutes, Claude Code summarizes it. When the user returns and gives the next instruction, it loads the summary instead of the entire conversation and continues.

Claude Code stores the conversation up to that point as a "prompt cache" for one hour and rereads it at a lower rate during that time. Once the cache expires, the next instruction rereads the whole conversation at the normal rate, which takes a large bite out of the usage limit. Because the summary is created before the cache expires, only the summary is reread after the user returns.

The feature was initially limited to the terminal version, but the Oct. 30 release extended it to the VS Code extension and the desktop app. The same release made it work for subscribers who have turned off usage data reporting (telemetry) and for use through other companies' clouds, such as Amazon Bedrock and Google Cloud's Vertex AI.

Summaries, however, lose detail. The developer blog dsebastien.net pointed out that rejected approaches, the exact text of error messages and conditions set at the start could drop out of the summary. The feature can be turned off in the settings.

Display of what usage was spent on

The company also expanded "/usage," which shows usage status. According to the changelog, the Oct. 14 release added warnings when the deepest reasoning setting, always-loaded instructions or large tool outputs are consuming a large share of the usage limit. The Oct. 20 release showed the cause and a one-line fix for each warning, and let users apply the fix with a single keystroke. The Oct. 26 release also listed rereads of the conversation caused by cache misses as items consuming a large share of the usage limit.

The Oct. 27 release renamed the five-hour usage limit from "session limit" to "5-hour limit." Anthropic on Oct. 30 (Oct. 31 Japan time) renamed its $200-a-month plan "Max 20x" to "Ultra" and removed the five-hour limit for Ultra users who have completed identity verification.

"ultrathink" deepens reasoning for one response, consuming more

The Oct. 26 release made writing "ultrathink" in an instruction deepen reasoning for that single response. Claude Code lets users choose from five reasoning levels, low, medium, high, xhigh and max, and the default for the latest "Opus 5.5," "Sonnet 5.5" and "Haiku 5.5" is medium. Writing ultrathink runs that response at xhigh. It applies to four models and only when using Anthropic's API directly.

Previously, ultrathink only added text to the instruction asking the model to think deeply and did not change the reasoning level setting. On Oct. 31, the VS Code extension was given the same behavior, and a setting to disable ultrathink was added.

Anthropic's official documentation describes xhigh as a setting that uses more tokens to reason deeply. Responses that use ultrathink consume more of the usage limit. In testing by Classmethod, the reasoning level stayed at the default in the non-interactive mode (claude -p).

Monthly API credits for top plans, not usable for Claude Code

On Oct. 7, the company began giving the top three subscription plans monthly credits usable with its developer API. The amounts match the monthly fees: $100 for the $100-a-month Max and $200 for the $200-a-month Ultra (then Max 20x). Team, for groups, adds up $20 per standard seat and $100 per premium seat, with a cap of $500 a month. The $20-a-month Pro, the free tier and Enterprise for large companies are not eligible.

The credits can be used for the API that calls Claude from users' own apps and AI agents, the Agent SDK development kit and "Managed Agents," which runs agents on Anthropic's servers. They cannot be used for Claude Code or for extra usage beyond the usage limit in the Claude apps. Usage limits for Claude Code and Claude are unchanged. Unused credits expire at the end of the billing period and do not carry over to the next month.

Team and Business add up per-seat amounts, capped at $500 a month.

OpenAI adopts the same arrangement 3 weeks later

OpenAI on Oct. 28 (Oct. 29 Japan time) announced that subscribers to its top-tier "ChatGPT Pro" plan and the business-focused "ChatGPT Business" would receive up to $500 a month in API credits. Pro plans at $100, $200 and $500 a month receive credits equal to their monthly fee. Business adds up $20 per standard seat and $100 per premium seat, capped at $500, the same amounts as Anthropic's design.

According to pasqualepillitteri.it, OpenAI's credits also cannot be used for its programming assistant "Codex" when accessed with an API key, and unused credits do not carry over. Google has, since January 2026, given Google Cloud credits of $100 a month to its paid "Google AI Ultra" plan and $10 a month to "Google AI Pro."

Most of Anthropic's revenue comes from usage-based fees for the API and other services. According to the registration statement for its listing (S-1), made public on Oct. 26, of revenue of $16.286 billion in January-June 2026, usage-based revenue was $13.064 billion and subscription revenue was $3.222 billion.

"mods" extensions and desktop app improvements

Mods, the first item listed in the post, were released on Oct. 1. They are programs users write in TypeScript or other languages to change Claude Code's interface and behavior. They can rewrite instructions, grant permissions for operations such as file access and add interface components. Because mods run on the computer with the same permissions as Claude Code, Anthropic asks users to install only those from trusted sources.

The Oct. 14 (Oct. 15 Japan time) update to the desktop app made it easier to run several tasks side by side. Users can start work from recently used folders, split work into child sessions, and see in a list which sessions are working and which are waiting for the user's reply.