Is Anthropic 'nerfing' Claude? Users increasingly report performance degradation as leaders push back

Users and developers are broadly অভিযোগing that Anthropic's Claude Opus 4.6 and Claude Code have become less capable, less reliable and more token-intensive, with some viral benchmark claims citing a drop from 83.3% to 68.3% accuracy and a No. 2 to No. 10 ranking. Anthropic denies secretly degrading the model, pointing instead to disclosed changes in effort defaults, thinking summaries, cache behavior and session limits, including a March 26 peak-hour limit change affecting about 7% of users and an April 7 default-effort shift for API-key and enterprise users. The story is negative for sentiment around Anthropic and may pressure customer trust and retention, but the evidence is mixed and not clearly indicative of a confirmed product regression.

Analysis

The market implication is not “Anthropic is definitely worse,” but that the monetization model for frontier coding assistants is more fragile than the sell-side assumes. If power users believe output quality is being traded off for token efficiency or quota management, enterprise retention becomes a trust problem, not a benchmark problem; that disproportionately hurts the vendors whose usage is deepest in long-horizon agentic workflows. The second-order winner is any model provider that can credibly market stable, reproducible behavior under load, even if raw benchmark scores are less impressive.

For AMD, the issue is reputational rather than fundamental in the near term. A senior AI hardware buyer publicly airing dissatisfaction with Claude Code matters because it raises the possibility that some enterprise inference spend gets reallocated away from “best model” usage and toward cheaper or more controllable stacks, which could modestly compress premium GPU mix at the margin if high-end reasoning tokens slow. The bigger medium-term risk is that capacity and product-tuning concerns push customers to multi-vendor routing, reducing concentration of spend with any one frontier lab and making inference demand less linear.

The main catalyst path is operational disclosure: if Anthropic clarifies defaults, cache behavior, and reasoning controls in a way that restores developer confidence, the controversy fades within days to weeks. If instead more users publish comparable logs showing degradation across longer time windows, the narrative can persist for months and affect renewal behavior in coding copilots and enterprise seats. The contrarian view is that some of this backlash may actually reflect successful cost optimization being misread as quality decay; that means the stock-level impact on Anthropic-adjacent exposure could be overstated if customers tolerate the tradeoff once pricing and controls become explicit.

Key Decisions for Investors

• Long RDDT / short AI-infrastructure basket for 2-6 weeks: Reddit should see elevated AI complaint traffic and engagement, while direct beneficiary names remain harder to monetize; use a tight stop if Anthropic resolves the issue publicly.
• Tactically underweight AMD for 1-2 months relative to broader semis: not a core demand thesis break, but the public complaint from a senior AMD AI executive increases the risk of internal procurement scrutiny around inference spend and vendor diversification.
• Buy short-dated call spreads on names positioned as stable enterprise AI alternatives over the next 1-3 months: the trade is for a trust-repair rotation if users defect from a single-vendor coding workflow to multi-model routing.
• Avoid chasing any sharp selloff in AI software on this headline; if anything, use weakness to add only where companies have explicit model-agnostic orchestration, because the more likely medium-term outcome is spend reallocation, not demand destruction.
• If Anthropic issues a transparent changelog or usage-policy clarification, fade the negative tape in AI software within 24-48 hours; this is a sentiment-driven event with fast mean reversion unless fresh evidence of quality decay emerges.