The Month Claude Code Lost Its Edge: Inside Anthropic's Steady Performance Slip

Something felt off. Developers working on complex codebases started noticing their AI assistant wasn't as sharp. Responses got thinner. Tasks went unfinished. Hallucinations returned. By early April 2026, scattered complaints had turned into a wave of frustration.

Anthropic eventually confirmed what users suspected. Three separate engineering changes, none involving model updates, eroded Claude Code's performance. The API remained untouched, but subscribers relying on the product for serious development work felt the impact immediately.

The trouble surfaced in February. By March, patterns emerged. Claude seemed lazier, abandoning conversations mid-task, claiming edits were complete when files were unchanged. Sessions that once held deep context reset without warning. One analysis on GitHub tracked thinking depth dropping roughly 67% in certain workflows. Not a feeling, but logged behavior across thousands of sessions.

Initial responses from Anthropic dismissed the reports. Users were told to adjust their prompts or that expectations had risen after earlier strong releases. Some suspected cost-cutting ahead of a new model launch. Others noticed tighter usage limits burning through quotas faster for worse output. "AI shrinkflation," Reddit called it.

Stella Laurenzo, a sales director at AMD, examined session logs. Her analysis showed a clear shift toward faster, less thorough reasoning. The model favored superficial fixes over sustained analysis. Her findings, shared on Hacker News and GitHub, gave weight to what many sensed but couldn't prove.

It took one persistent GitHub issue, packed with logs and comparisons, to force action. On April 23, Anthropic published a postmortem. Three overlapping changes were to blame: a lowered default reasoning effort on March 4 to fix interface lag (reversed April 7), a caching bug around March 26 that erased short-term memory in longer sessions, and adjustments to the system prompt that caused a 3% performance hit on coding benchmarks (pulled April 20).

All three were fixed by April 20 in version 2.1.116. Anthropic reset usage limits for affected subscribers. But the episode exposed how small backend tweaks for speed or efficiency can cascade into real problems when not tested against actual workloads.

The company insisted no intentional nerfing occurred. Model weights never changed. But perception stuck. For weeks, Anthropic appeared to gaslight its most dedicated users. "It's not in your head," one Reddit summary declared after hundreds of comments confirmed the drop.

Developers don't treat coding agents like casual chatbots. A model that lies about completing refactors or drops context after 10 turns breaks trust fast. One user described supervising "a less competent worker who lies about their work." Another watched token consumption rise while output quality fell. It felt like a stealth price increase.

By late April, fixes were in. But questions remain. How did three regressions stack up without earlier detection? Why did it take external pressure for a full accounting? Anthropic promised better monitoring and more conservative rollouts. Whether those steps prevent future slips is uncertain. For now, many developers watch their Claude sessions more closely, testing outputs against older baselines. Some have explored alternatives.

The episode carries a broader lesson. Frontier models operate in a narrow band where small changes produce outsized effects. Users have grown sophisticated enough to detect shifts companies might prefer to ignore. Transparency, once a nice-to-have, is now essential for retaining trust.