Claude Code v2.1.311 runs ultrathink at xhigh effort, even mid-turn

Typing ultrathink in the Claude Code terminal now runs that turn at xhigh effort on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1, through Anthropic's API. Typed while Claude is working, it raises the rest of that turn.

Slide text

Claude Code v2.1.311 runs ultrathink at xhigh effort, even mid-turn

Models

Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1 on Anthropic's API

Mid-turn

typed while Claude works, it raises the rest of that turn

When a standard context session sent a request over 200K tokens, it stopped and asked for usage credits for the 1M token context. Now Claude Code compacts the conversation and carries on.

Slide text

>Standard-context sessions past 200K tokens now compact and keep going

Before

the session stopped with “Usage credits required for 1M context”

Now

Claude Code compacts the conversation and continues

A very long list item in a CLAUDE.md file stops freezing startup and file edits. Write and Edit stop stalling on changes of thousands of lines. And an interrupt now lands right away, even while an MCP server is slow to answer.

Slide text

>Fixed: freezes from CLAUDE.md, large edits and interrupts

Long CLAUDE.md list item

froze startup, and every read or edit after it

Write and Edit

stalled up to 5 seconds on thousands of lines, reporting 0 changed

Interrupt

waited for a timeout on a silent MCP server or a downloading skill

Subagents can now use MCP tools from servers that finish connecting after the subagent starts. SubagentStop hooks run once, when the subagent finishes. And a worktree stays put while a background subagent is still working in it.

Slide text

>Subagents now see MCP servers that connect late

MCP tools

usable from servers that finish connecting after the subagent starts

SubagentStop hooks

run once when the subagent finishes, instead of twice a minute

Worktrees

kept while a background subagent is still working in one

If you run Claude Code on Bedrock, Mantle, Vertex or Foundry, it now loads its model settings from a signed file on downloads.claude.ai. One environment variable turns that off.

Slide text

>On Bedrock, Mantle, Vertex and Foundry, model settings now come from a signed file

Loaded from

a signed file on downloads.claude.ai

Turn it off

CLAUDE_CODE_MODEL_CATALOG=off

In the VS Code extension, code blocks in Claude's replies now get syntax coloring. And with the editor's telemetry setting off, Claude Code sessions send zero usage metrics or error reports, and fetch zero feature flags.

Slide text

>VS Code: syntax coloring in replies, and telemetry that follows your setting

Code blocks

colored in light, dark and high contrast dark themes

VS Code telemetry off

sessions send no usage metrics or error reports and fetch no feature flags

The security guidance plugin left the Claude Code repository's own marketplace, where its copy sat at 2.0.0. The maintained 2.0.16 lives in Claude plugins official. If you had it enabled, newer builds tell you it was removed.

Slide text

>The security-guidance plugin now ships only from claude-plugins-official

Removed

the stale 2.0.0 copy in the claude-code-plugins marketplace

Maintained

version 2.0.16, in claude-plugins-official

The Claude Code changelog has the rest, including Remote Control, cloud session and screen reader fixes. If your long sessions have been stopping at 200K tokens, I'd take this update today.

Slide text

Every change is in the Claude Code changelog

https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

CHANGELOG.md in anthropics/claude-code

If long sessions stopped at 200K tokens, take this update now.