Claude users complain of degradation

Anthropic’s Claude users spent this week arguing that the chatbot has become less reliable, while the company’s own teams said recent product changes altered how it behaves. (techzine.eu) The complaints center on Claude Opus 4.6 and Claude Code, Anthropic’s coding product, with users posting on GitHub, X, and Reddit that the system crashes more on complex tasks, stops early, and uses more tokens without better answers. VentureBeat reported the backlash on April 13, 2026, and Techzine said the debate intensified on April 14. (venturebeat.com) (techzine.eu) One widely shared complaint came from Stella Laurenzo of Advanced Micro Devices, who said in a GitHub issue filed April 2, 2026 that she analyzed 6,852 Claude Code session files, 17,871 thinking blocks, and 234,760 tool calls. She said the logs showed shallower reasoning and more abandoned or simplistic fixes starting in February. (venturebeat.com) (techzine.eu) Large language models are prediction engines that spend computing power, measured partly through token use, to work through a prompt step by step. Fortune reported Anthropic reduced Claude’s default “effort” setting to “medium” to cut token consumption, which changes how much work the model does by default before answering. (finance.yahoo.com) Anthropic executives have disputed the strongest accusations. Boris Cherny, who leads Claude Code, said recent shifts came from product choices such as interface changes and default settings, and Thariq Shihipar of the Claude Code team said the company does not weaken models to manage peak demand. (techzine.eu) Other developers have also pushed back on the benchmark claims. Techzine cited researcher Paul Calcraft saying the tests being compared use different datasets, which makes a sharp before-and-after verdict harder to prove from public numbers alone. (techzine.eu) The dispute lands as Anthropic presents Claude less as one chatbot than as a lineup of models with different tradeoffs. In its current documentation, Anthropic says Claude is a “family” of models and lists Claude Opus 4.6, Sonnet 4.6, and Haiku 4.5 with different speed, price, and context-window limits. (platform.claude.com) Those docs say Opus 4.6 is Anthropic’s most capable broadly available model for coding and reasoning, Sonnet 4.6 is the speed-and-intelligence middle tier, and Haiku 4.5 is the fastest tier. Anthropic also says models with the same snapshot date are identical across platforms and “do not change,” a promise aimed at developers who need stable behavior. (platform.claude.com) Anthropic’s public system-card page shows how quickly that lineup has changed, with cards for Sonnet 4 and Opus 4 in May 2025, Opus 4.1 in August 2025, Sonnet 4.5 in September 2025, Haiku 4.5 in October 2025, Opus 4.5 in November 2025, and Opus 4.6 and Sonnet 4.6 in February 2026. That release pace has given developers more options, but it has also made it harder to tell whether a bad result comes from a new model, a new default, or a new interface. (anthropic.com)