An update on recent Claude Code quality reports

2026-04-23 · Source: Anthropic Engineering Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Anthropic identified and resolved three distinct issues that led to recent reports of degraded code quality in Claude Code, Claude Agent SDK, and Claude Cowork, affecting Sonnet 4.6, Opus 4.6, and Opus 4.7. The first issue, from March 4, involved changing Claude Code's default reasoning effort from `high` to `medium` to reduce latency, which was reverted on April 7 due to user preference for higher intelligence. The second, on March 26, was a bug in a caching optimization that caused Claude to continuously clear older thinking from idle sessions, leading to forgetfulness and repetition; this was fixed on April 10. The third issue, introduced on April 16, was a system prompt instruction to reduce verbosity that inadvertently harmed coding quality and was reverted on April 20. All issues were resolved by April 20 (v2.1.116), and Anthropic is resetting usage limits for all subscribers as of April 23, 2026.

Key takeaway

For engineering leaders overseeing AI product development, this incident underscores the need for robust change management and testing protocols. Your teams should implement gradual rollouts, expand internal testing with public builds, and enhance automated code review with broader context. Prioritize user feedback channels to detect subtle degradations that internal metrics might miss, especially when balancing performance tradeoffs like latency versus intelligence. This proactive approach minimizes unexpected quality regressions.

Key insights

Three distinct changes caused Claude Code's perceived degradation, highlighting the complexity of LLM product management.

Principles

• Prioritize intelligence over latency for complex tasks.
• Thoroughly test prompt changes across broad eval suites.
• User feedback is critical for identifying subtle degradations.

Method

Anthropic traced degradation reports to specific changes, performed ablations on system prompts, and used an improved Code Review tool to identify and fix bugs.

In practice

• Implement gradual rollouts for significant model changes.
• Use internal dogfooding with public builds.
• Expand context for automated code review tools.

Topics

• Claude Code
• Model Degradation
• Reasoning Effort
• Caching Optimization
• System Prompt Changes

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• An update on recent Claude Code quality reports — Harness-level bugs, not model issues, caused Claude Code's recent performance degradation.
• Claude Opus 4.7 felt dumber than 4.6. Then I added five lines to my personalisation. — Claude Opus 4.7 offers deeper analysis than 4.6, but requires explicit prompting to reveal its full rigor.
• v2.1.89 — Claude Code updates focus on performance, stability, and expanded capabilities for developers.
• Level Up Your Claude Code with This CLAUDE.md — A "claude.md" file defines an AI assistant's engineering workflow, ensuring consistent, disciplined code generation.
• Mystery solved: Anthropic reveals changes to Claude's harnesses and operating instructions likely caused degradation — Product-layer changes, not model degradation, caused Claude's perceived performance issues.
• 23 Tips for Smart Claude Code Token Saving and Workflow Optimization — Proactive context management and strategic tool usage are crucial for controlling Claude Code token costs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Anthropic Engineering Blog.