Claude Code 2.1.321: What the Biggest Maintenance Drop of 2026 Means for Your Workflow
Claude Code 2.1.321 introduces Claude Fable 5.2 as the new default model with a 1 million token context window, alongside over 40 bug fixes and key behavioural changes to Remote Control, mods, and permission handling. Non-technical teams — including legal and finance professionals in India managing large document sets — will feel the most direct impact from the expanded context capacity and tightened security fixes.
What Just Shipped
According to the Claude Code changelog for version 2.1.321, Anthropic has pushed one of its most comprehensive maintenance and feature releases for Claude Code to date. The update spans new model additions, dozens of behavioural changes, and over forty bug fixes — touching everything from how Claude handles images in long sessions to how it installs itself on Windows. If you use Claude Code even occasionally, several of these changes will affect your day-to-day experience directly.
The Headline: Claude Fable 5.2 Is Now the Default
The most consequential addition in this release is Claude Fable 5.2 (claude-fable-5-2). The Claude Code changelog confirms it is now the default Fable model, with a 1 million token context window and pricing set at \$8 per million input tokens and \$40 per million output tokens, with cache reads available at \$0.20 per million tokens. Converting to INR at approximately ₹85 per dollar, that works out to roughly ₹680 per million input tokens and ₹3,400 per million output tokens — positioning it as a high-capability model aimed at power users and teams handling large codebases or document sets.
If you previously saved a preference for Fable 5.1, the changelog notes that those saved picks will automatically migrate to the generic fable: selector on Pro and Max plans immediately, and on Team and Enterprise plans once your organisation enables it. If you specifically need to stay on Fable 5.1, the changelog instructs users to run /model claude-fable-5-1 to switch back manually.
What Changed in How Claude Works Day to Day
Remote Control Gets Smarter — and More Restricted
The --settings flag has been added to claude remote-control, allowing a settings file to be applied to every session the remote control starts. The changelog notes an important constraint: if that settings file changes, no new session will start until you restart claude remote-control. This is a deliberate safety boundary, not a bug.
Separately, Remote Control sessions now keep the folder they were started in — a connected client can no longer move a session to a different folder mid-run. This prevents a class of subtle errors where tooling or hooks might operate on the wrong working directory.
Mods Now Turn On Automatically in Auto and Bypass Modes
The changelog records a meaningful behavioural shift: mods that Claude writes will now activate automatically in auto and bypass modes, even when managed settings like allowUnsandboxedCommands: false or allowManagedDomainsOnly are in place. In dontAsk mode, mods will not load. Additionally, a mod’s own tool calls — $.tool.call and $.mcp.call — will no longer trigger permission prompts, even under custom ask rules, though deny rules and organisation policy still apply.
Desktop Code Tab Sessions Can Send More Messages
A quiet but practical improvement: the number of messages a Claude Desktop Code tab session can send to your other sessions before waiting for you to type again has increased from 10 to 20. For users who rely on background agents coordinating across sessions, this reduces interruptions during complex, multi-step tasks.
Fast Mode Labels Are Clearer
The changelog notes that fast mode labels in /fast, /config, and /model now name your current model when it supports fast mode, and the “(Opus only)” label has been removed from the VS Code row. This is a cosmetic change but reduces confusion for users who were unsure whether fast mode was available to them.
A Scenario: A Legal Operations Team in Pune
Consider a legal operations team at a mid-sized firm in Pune that uses Claude Code to draft contract summaries, flag clause anomalies, and cross-reference regulatory documents. Their contracts often run to hundreds of pages, and they frequently need Claude to hold the entire document set — plus relevant case notes — in memory during a single session.
With Claude Fable 5.2’s 1 million token context window now set as the default, the team no longer needs to manually break large document batches into smaller chunks and pass them across multiple sessions. The changelog’s improvement to how Claude handles sessions with more images than fit in one request is also relevant here: Claude is now told whether it has already seen a dropped image, and repeat reads of an unchanged image are limited — meaning the system is smarter about not wasting context on content it has already processed.
For the firm’s IT coordinator managing Remote Control sessions, the new --settings flag means a standard configuration — approved tool permissions, organisation policy, preferred model — can be applied consistently to every session without manually reconfiguring each one. The restriction that settings file changes require a restart is a tradeoff the team will need to plan around, but it also prevents accidental mid-session policy drift.
Security Fixes Worth Knowing About
The changelog lists several fixes that carry real security weight, even if they sound technical:
- MCP OAuth fix: A bug was fixed where MCP OAuth requests were sending secrets such as refresh tokens to another host when a server redirected them. If your team uses MCP servers with OAuth authentication, this fix closes a meaningful credential exposure vector.
- Sandbox escape fix: A case where a read-only Bash command ran outside the sandbox without a permission prompt has been resolved.
- File path checks: Deny rules and protected-file prompts were not applying to the file that a
cpormvinto a directory writes. This has been fixed. - Account email check tightened: The account email check now refuses only a shell command that holds your email and names a User-Agent, without asking — so other commands and scheduled runs no longer stop at its prompt unnecessarily.
None of these require action from non-technical users, but they are worth knowing about if you manage Claude Code deployments for a team.
Limitations and Honest Tradeoffs
Several things in this release come with constraints that the changelog states plainly.
What to Watch Next
The Claude Code changelog for 2.1.321 is notably dense — the sheer number of fixes suggests Anthropic is in a phase of hardening the platform after rapid feature expansion. The addition of Ultra alongside Pro and Max in session feedback messages and the allowedProviders setting description hints that Claude Ultra plan support is being woven more deliberately into the Code product’s infrastructure.
If you are evaluating whether to upgrade your team’s Claude plan, the clearest signal from this release is that Fable 5.2’s 1 million token context window is now the baseline for the Fable tier. Teams handling large Indian legal, financial, or regulatory document sets — where document volume routinely strains smaller context windows — have a concrete reason to revisit what their current plan supports and whether the per-token cost at ₹680 per million input tokens fits within their workflow economics.
