Claude Code’s late-October updates touch everyday development work, from long sessions to reasoning controls. Anthropic’s release notes clarify the versions and conditions behind several of them.
More reasoning effort for a turn
In the October 26 release notes for 2.1.311, Anthropic says ultrathink in the interactive terminal runs a turn at xhigh effort for specified supported models through Anthropic’s API. That wording does not make it a universal switch for every model and provider.
The October 31 version 2.1.316 added an environment variable to disable the keyword and a separate way to set default subagent effort when nothing else specifies it.
Long commands and usage visibility
According to the 2.1.311 notes, when a shell command reaches its timeout the result shows output so far and says the command continues in the background. The release also fixed a case where large prompt-cache misses were absent from the /usage breakdown.
Automatic compaction makes room by summarizing part of a long conversation. The release includes fixes around that process, but does not promise a fixed token saving for every long task.
