What Was Announced
Anthropic shipped a series of significant Claude Code updates throughout March 2026, documented in the official changelog. The changes span from major model capability expansions (a 1M token context window) to new input modalities (voice), developer workflow improvements (new slash commands), and advanced scripting capabilities. This is one of the most feature-dense months Claude Code has seen.
Key Features and Changes
Voice Mode (/voice)
The most surprising addition is voice input, activated via the /voice command. It uses a push-to-talk model — hold the spacebar to speak, release to send. The feature launched with 10 languages and expanded to 20 by the end of March. For developers who dictate faster than they type, or who are debugging while their hands are occupied, this unlocks a genuinely different interaction model.
1 Million Token Context Window (Opus 4.6)
Starting from version 2.1.75, Opus 4.6 supports a 1 million token context window by default for Max, Team, and Enterprise plan users — a 5× increase from the previous 200K limit. In practice this means entire codebases, including all source files, test suites, and documentation, can fit in context simultaneously. Previously, Claude Code would compact or truncate context aggressively on large repos; that concern is largely eliminated for most projects.
New Slash Commands
/effort— Sets the model's reasoning effort level. Useful for trading off response speed against thoroughness on complex tasks./simplify— Triggers a code quality review pass on recently changed code, checking for reuse opportunities, unnecessary complexity, and inefficiency./batch— Runs a prompt or slash command against multiple targets in parallel, useful for applying the same operation across several files or modules.
Smarter /context Command
The /context command was upgraded to be proactive rather than passive. It now identifies which tools and memory entries are consuming the most context budget, warns when approaching compaction limits, and suggests specific actions (e.g. "clear this memory file", "detach this MCP server").
Scripting and Automation Features
--channelspermission relay — Channel servers that declare the permission capability can now forward tool-approval prompts to your phone, enabling remote approval of sensitive operations.--bareflag — Skips hooks, LSP, plugin sync, and skill directory walks for scripted-pcalls, making Claude Code more predictable in CI pipelines.autoMemoryDirectorysetting — Configure a custom path for the automatic memory directory, useful for shared team setups.
Why It Matters for Developers
The 1M context window fundamentally changes how Claude Code handles large monorepos and multi-module projects. Tasks like "find all usages of this deprecated API across the codebase and replace them" or "understand the full data flow from HTTP request to database query" are now tractable in a single session without manually managing what stays in context.
The /batch command is a sleeper hit — it enables parallelising repetitive AI-assisted tasks (e.g. generating tests for 20 service classes) that would otherwise require manual orchestration.
Source
Read the original article — Claude Code Docs
