Full Breakdown
Anthropic’s Claude Code Faces Performance Crisis After Engineering Missteps
4/25/2026, 9:47:07 AM
Claude Code’s Performance Decline Sparks User Backlash
Anthropic confirmed that three engineering changes introduced between March 4 and April 16 2026 degraded Claude Code’s output, prompting a wave of complaints, subscription cancellations, and accusations of “gaslighting.” All three issues were fixed by April 20, and usage limits were reset on April 23.
Background: Claude Code’s Role in Anthropic’s Growth
Claude Code, launched in early 2025, quickly became a flagship AI coding assistant for solo developers and enterprise teams, helping lift Anthropic’s annual recurring-revenue run rate to $30 billion as the firm prepared for an IPO.
Timeline of the Three Missteps
March 4: reasoning effort lowered from high to medium, cutting latency but reducing intelligence. March 26: a cache-optimization bug cleared reasoning turn, causing forgetfulness and faster usage drain. April 16: a prompt limited responses to 25 words between tool calls, hurting quality. Anthropic reverted the effort setting on April 7, fixed the cache bug on April 10, and rolled back the prompt on April 20; usage limits were reset.
Data & Statistics
Similarweb reported 20 million Claude app users by February, with growth doubling in March. Veracode found vulnerabilities in 52 % of coding tasks for Claude Opus 4.7, versus about 30 % for OpenAI’s models. TrustedSec measured a 47 % drop in Claude’s code-quality metrics during the period.
Official Statements & Responses
Anthropic’s post-mortem said the three problems were “specific technical causes” and that API and inference layer were never degraded. It pledged tighter prompt controls, a feedback channel @ClaudeDevs, and reset usage limits on April 23. An OpenAI memo, reported by CNBC, called Anthropic’s compute strategy a “strategic misstep” and claimed the company was operating on a “meaningfully smaller curve” than rivals; Anthropic declined to comment.
Criticism, Opposition, and On-the-Ground Reports
Users accused Anthropic of “gaslighting” after a test removed Claude Code from the $20-a-month Pro plan for about 2 % of new sign-ups. Head of growth called it a usage-pattern experiment, while analysts warned that a shift to enterprise pricing could triple costs for users. Developers on X and Reddit called the tool “lazy,” “ignorant,” and “unusable for complex engineering,” prompting cancellations and migration to OpenAI’s Codex and GPT-5.5.
Conflicting Reports & Gaps
Anthropic has not disclosed compute capacity figures or the exact number of cancelled subscriptions, while OpenAI’s memo suggests a compute shortfall without quantifying the gap. Neither company released the financial impact of the episode.
Verbatim Quotes
- “Claude has regressed to the point it cannot be trusted to perform complex engineering,” — Stella Laurenzo, senior director, AMD AI Group
- “I’m glad they are trying to address this, but a month to get this out is crummy.” — Dave Kennedy, CEO, TrustedSec
- “Claude was dumbified on March 4, just when we noticed!” — @levelsio, X user
- “Claude Code with Opus 4.6 was so dumb today I finally had to write my own code again.” — X user, anonymous
What’s Next
Anthropic and OpenAI are both gearing up for IPOs later in 2026. OpenAI’s GPT-5.5 launch may draw users who left Claude Code, while Anthropic plans a cautious rollout of future Claude models with tighter prompt controls and expanded public feedback channels.
