3Claude Code Release Notes·
Claude Code 2.1.201: Sonnet 5 sessions drop mid-conversation system role for harness reminders
Claude Code release 2.1.201 removes the use of the mid-conversation system role for harness reminders in Claude Sonnet 5 sessions. This is a narrow behavioral change to how the Claude Code agent harness communicates with the model during sessions. The change likely reflects improvements in Sonnet 5's ability to maintain context without mid-session system-role injections.
Related guides (3)
Related events (8)
Claude Code 2.1.216 release: quadratic slowdown fix, agent session restoration, and broad bug fixes
Anthropic released Claude Code version 2.1.216, a maintenance update addressing a significant performance regression where message normalization cost grew quadratically with conversation length, causing multi-second stalls in long sessions. The release also fixes agent session state restoration (prompt and tool restrictions now persist across resumes), OAuth token expiry mid-session, and numerous edge cases in worktree isolation, background subagent lifecycle, and MCP re-authentication. Additional improvements cover UI rendering, telemetry accuracy, PowerShell/Bash command parsing, and cloud session resilience on container restart.
Anthropic Releases Claude Sonnet 4.6 with 1M Token Context, Improved Computer Use, and Coding Capabilities
Anthropic has released Claude Sonnet 4.6, positioned as a major upgrade over Sonnet 4.5 with improvements across coding, computer use, long-context reasoning, and agent planning. The model features a 1M token context window in beta and is now the default on claude.ai Free and Pro plans at unchanged pricing ($3/$15 per million tokens). Notably, users preferred Sonnet 4.6 over the prior Opus 4.5 frontier model 59% of the time in coding tasks, and the model shows significant gains on OSWorld computer-use benchmarks alongside improved prompt injection resistance. Safety evaluations found no major alignment concerns and rated it as safe or safer than prior Claude models.
Claude Code 2.1.176 release: session language, Bedrock caching, model allowlist fixes, Remote Control improvements
Anthropic released Claude Code version 2.1.176 with a mix of new features and bug fixes. Notable additions include session titles generated in the conversation language, a new footerLinksRegexes setting, and improved AWS Bedrock credential caching. Key fixes address model allowlist enforcement (alias models can no longer bypass blocked-model restrictions), auto-mode fallback for organizations without Opus 4.8 enabled, hook path pattern matching, and numerous Remote Control and background-session reliability issues across Linux, Windows, and tmux environments.
Claude Code 2.1.210 release: bug fixes, security hardening, and agent workflow improvements
Anthropic released Claude Code version 2.1.210, a substantial patch addressing over 30 bugs and adding several improvements to the agentic coding tool. Notable fixes include hardening against indirect prompt injection via subagent-read content, correcting git worktree isolation for subagents, fixing hook callback timeouts that caused unattended sessions to stall, and resolving MCP server teardown during mid-session re-syncs. The release also improves auto-mode permission classification by defaulting to Sonnet 5 for external sessions and adds UX enhancements for the multi-agent dashboard.
Claude Code 2.1.212 release: /fork background sessions, runaway-loop caps, and ~40 bug fixes
Anthropic shipped Claude Code version 2.1.212 with several notable behavioral changes: /fork now copies conversations into independent background sessions (previously launched in-session subagents), and new per-session caps on WebSearch calls (default 200) and subagent spawns (default 200) guard against runaway loops. MCP tool calls exceeding 2 minutes are automatically backgrounded to keep sessions responsive. The release also includes roughly 40 bug fixes spanning plan-mode permission bypasses, SIGTERM process-tree orphaning, Windows PowerShell compatibility, /ultrareview edge cases, OpenTelemetry export failures, and prompt caching improvements for LLM gateways.
Claude Code 2.1.214 release: security fixes, EndConversation tool, and extensive bug fixes
Anthropic released Claude Code version 2.1.214, a substantial patch addressing multiple permission-check security vulnerabilities including a Windows PowerShell 5.1 bypass, file-descriptor redirect bypass, and path-traversal issues in allow rules. The release adds the EndConversation tool enabling Claude to terminate sessions with abusive users or jailbreak attempts, improved OpenTelemetry observability attributes, and docker/Podman permission prompts. Dozens of bug fixes address Windows-specific issues, background session lifecycle problems, memory management, and streaming reliability.
Claude Code 2.1.200 release: permission mode defaults to Manual, background agent fixes
Anthropic released Claude Code version 2.1.200 with a notable default behavior change: the permission mode now defaults to 'Manual' across CLI, VS Code, and JetBrains integrations, replacing the previous auto-continue behavior for AskUserQuestion dialogs. The release includes extensive fixes for background agent reliability issues including daemon lock conflicts, session stalls after sleep/wake cycles, rate-limit handling, and MCP server configuration errors. Accessibility improvements for screen readers and terminal rendering fixes under tmux 3.4+ are also included.
Anthropic Releases Claude Sonnet 4.5: Top Coding and Computer-Use Model with Agent SDK
Anthropic has released Claude Sonnet 4.5, claiming it is the best coding model and strongest model for building complex agents, with a 61.4% score on OSWorld (up from 42.2% for Sonnet 4) and state-of-the-art performance on SWE-bench Verified. The release is accompanied by major product upgrades including checkpoints in Claude Code, a native VS Code extension, a Claude Agent SDK giving developers access to the same infrastructure powering Claude Code, and new context editing and memory tools in the Claude API. Pricing is unchanged from Sonnet 4 at $3/$15 per million input/output tokens. Early enterprise customers including Cursor, GitHub Copilot, Devin, Canva, and Figma report significant gains in coding, agentic, and long-context tasks.