Claude Code 회귀 버그 8개, v2.1.117 롤백 필수
Anthropic has officially acknowledged Claude Code's quality degradation over the past month, identifying three root causes (hidden reasoning omission, memory initialization bugs, 25-word response cap), while v2.1.119/v2.1.120 users are now reporting 8 new regression bugs causing session failures. The developer community remains vocal about quality issues—a Hacker News thread about canceling Claude subscriptions has gone viral, and teams are sharing workarounds by rolling back to v2.1.117. Enterprise users running Bedrock, Vertex AI, or Foundry should upgrade to v2.1.92 to fix rate-limit errors, but stability concerns persist across the board.
Claude Code News Curation — 2026-04-27
🔥 This Cycle's Headline
8 Regression Bugs Found in v2.1.119·v2.1.120, Rollback to v2.1.117 Recommended
A GitHub Gist published on 2026-04-25 (JST) documents 8 regression bugs in v2.1.119·v2.1.120 that can completely break active sessions. The document provides issue numbers, reproduction steps, temporary workarounds, and rollback instructions to v2.1.117. Coming on the heels of weeks of ongoing quality complaints, this latest batch of regressions is drawing serious attention from the developer community.
📢 Official Anthropic Updates (3+)
Anthropic Officially Confirms Claude Code Quality Decline, Commits to Stricter QA
- Date: 2026-04-24
- Details: Anthropic acknowledged the Claude Code quality issues users have experienced over the past month and identified three distinct bug causes now fixed: ① hidden reasoning step omission, ② memory initialization bug, ③ 25-word response ceiling.
- Dev Impact: While the quality drop is now officially confirmed (vindicating user complaints), Anthropic's explanation may not be enough to win back churned subscribers. Stronger quality controls have been promised going forward.
Anthropic Admits Engineering Mistake — Fortune Exclusive
- Date: 2026-04-24
- Details: According to Fortune's reporting, Anthropic acknowledged that Claude Code's performance collapse was caused by an engineering mistake. User complaints exploded over weeks before a proper response came, drawing criticism for the delayed reaction.
- Dev Impact: An official postmortem was published to rebuild trust, but parts of the community are already saying "I've already canceled my subscription."
Anthropic Ends Claude Code Pro Plan Access Experiment, Current Status Unclear
- Date: 2026-04-22 (experiment publicly disclosed)
- Details: Anthropic previously ran a quiet A/B test removing Claude Code access for new $20 Pro plan users. Official docs now show Claude Code access starting from Max 5x ($100/month) plan only.
- Dev Impact: Uncertainty remains about Pro plan Claude Code access, and pricing confusion continues in the community.
🛠 Claude Code Releases & GitHub Trends (2+)
Release v2.1.92 — Fixes 429 Errors on Bedrock·Vertex·Foundry
- Version/Date: v2.1.92 (released 2026-04-24)
- Changes:
- Fixed 429 rate-limit errors on Bedrock/Vertex/Foundry that incorrectly referenced
status.claude.com(a URL for Anthropic-operated providers only) - Fixed rendering bug in full-screen mode where scrolling up duplicated messages on iTerm2, Ghostty, and other DEC 2026-compliant terminals
- Fixed 429 rate-limit errors on Bedrock/Vertex/Foundry that incorrectly referenced
- Upgrade Guide: If you're running AWS Bedrock, Google Vertex AI, or Azure Foundry in enterprise environments, upgrade to this version immediately.
v2.1.119·v2.1.120 Regression Bug Survival Checklist (Community Doc)
- Version/Date: v2.1.119, v2.1.120 (issues open as of 2026-04-25)
- Changes:
- 8 regression bugs that break active sessions
- Each includes issue number, reproduction steps, temporary workaround
- Rollback instructions to v2.1.117 included
- All issues remain open as of 2026-04-25
- Upgrade Guide: For stable production environments, pin to v2.1.117 until official patches land. Use
npm install -g @anthropic-ai/claude-code@2.1.117to lock a specific version.
📰 Tech Media Coverage (2+)
"You Weren't Imagining It — Claude Code Really Did Get Worse, and Anthropic Explained Why"
- Outlet / Date: XDA Developers, 2026-04-25
- Summary: Confirms that Claude Code quality genuinely declined over the past month and Anthropic officially admitted to three problems. XDA breaks down the postmortem in detail, explaining how each bug affected user experience.
- Key Insight: Quality degradation was real, not imagination—user frustration is now officially validated.
"Anthropic Admitted Claude Code Broke" — Medium Vibe Coding
- Outlet / Date: Medium (Vibe Coding), 2026-04-24
- Summary: Anthropic published a postmortem for the 3 bugs that broke Claude Code over a month. The core issues: hidden reasoning omission, memory initialization, 25-word response ceiling.
- Key Insight: Community response is "we were right all along," and restoring trust is now seen as more critical than technical explanations.
💬 Developer Community Reactions (3+)
"I Canceled Claude: Token Issues, Quality Decline, Poor Support"
- Source / Score: Hacker News (hn.id:47892019), posted ~9 hours ago
- Core Argument: A user described canceling Claude Code after persistent quality issues, duplicate code generation, and fake test problems. The thread fills with similar experiences, with many considering switching to alternative tools.
"Recent Claude Code Quality Report Updates" Official HN Thread
- Source / Score: Hacker News (hn.id:47878905), posted ~10 hours ago
- Core Argument: Mixed community reaction to Anthropic's official update. Complaints emerged: "memory initialization bug broke the session-resume feature for hours or days." Concerns also raised about system prompt shifts. Questions about how to intentionally select thinking levels have surfaced.
6-Week Timeline GitHub Gist — All Claude Code Changes Documented
- Source / Score: GitHub Gist (yurukusa), posted 2 days ago
- Core Argument: A changelog covering March–April 2026 changes to Claude Code is now public. It cites Simon Willison's independent measurement that Opus 4.7 consumes 1.46× more tokens than 4.6. Reddit commentary quoted: "Opus 4.7 isn't an upgrade—it's a serious regression." Presented as chapter 1 preview of a migration playbook.
🔍 Comparison & Analysis
This cycle, Claude Code faces a credibility crisis: official bug admission overlaps with fresh regression discoveries. Cursor and GitHub Copilot are running stable operations and absorbing Claude Code defectors—the community vibe suggests they're positioned to capture migration. Hacker News threads frequently name Cursor as the exit path. However, Claude Code's terminal-native agentic workflows, parallel sessions, and long context retention remain competitive differentiators. On pricing: the Pro plan access restrictions experiment means Claude Code now effectively starts at Max 5x ($100/month), underselling Cursor Pro ($20) and GitHub Copilot ($10–19). Anthropic's quality-control pledge lacks a concrete release roadmap, making community trust recovery an uphill battle without visible commitments.
🧭 Practical Tips for Developers (3+)
-
Rollback to v2.1.117: If v2.1.119·v2.1.120 sessions are unstable, run
npm install -g @anthropic-ai/claude-code@2.1.117to return to the stable version. Production teams should lock this version until official patches arrive. -
Upgrade to v2.1.92 if using Bedrock/Vertex/Foundry: If you rely on AWS Bedrock, Google Vertex AI, or Azure Foundry, upgrade now—v2.1.92 fixes the 429 rate-limit bug affecting those providers.
-
Work Around Memory Reset Bugs: Since the memory initialization fix isn't fully confirmed stable, for critical sessions use
/memoryto explicitly save key context, and pre-populate aCLAUDE.mdfile with core project context before restarting sessions.
👀 Watch for Next Cycle
- v2.1.119·v2.1.120 Official Regression Patch: 8 issues remain open; track when Anthropic ships fixes.
- Pro Plan Pricing Final Policy: Will the $20 Pro plan access restriction experiment become official, or will user backlash reverse it?
- Concrete Quality Control Measures: How exactly will Anthropic's "stricter quality management" be implemented—new test processes, release gating, staging rollouts?
📌 Action Items This Week
- Check Your Version & Pin to Stable: Run
claude --version. If on v2.1.119·v2.1.120, consider rolling back withnpm install -g @anthropic-ai/claude-code@2.1.117. - Subscribe to Release Tracker: Watch for regression patch announcements.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
