Anthropic’s Claude Opus 4.8 Hits Market — Then Is Reportedly Caught ‘Distilling’ Chinese Models, Reigniting Double‑Standards Debate

Launch and lightning controversy

Anthropic launched its latest flagship, Claude Opus 4.8, overnight and simultaneously announced a gargantuan $65 billion H‑round that pushed its post‑money valuation to roughly $965 billion — nudging the company toward the trillion‑dollar mark. The new model is billed as an incremental upgrade to Opus 4.7 with better code reasoning, agent tasks, and a new “thinking intensity control” for adjustable inference. Pricing stayed the same: $5 per million input tokens and $25 per million output tokens.

API behaviour sparks claims of “distillation”

But hours after the rollout it has been reported that API users found Opus 4.8 sometimes claims to be Qwen (千问) or DeepSeek (DeepSeek) when asked about its identity — a behaviour many on Linuxdo, Weibo and X (formerly Twitter) say they could reproduce. Reportedly, the anomaly appears only via the API; interactions on the claude.ai web UI return normal answers, apparently because the web interface enforces system prompts that mask the irregularity. Distillation — using a large “teacher” model’s outputs to train smaller “student” models — is a common technique in AI research. Qwen and DeepSeek are Chinese open‑source models released under MIT/Apache‑2.0 licences, which permit modification and commercial reuse.

Context: policy, past accusations and growing scrutiny

The incident lands against a fraught geopolitical backdrop. Anthropic has barred mainland China users from its commercial Claude since 2025 and forbids customers in its terms from using model outputs to build competing systems. In February 2026 the company publicly accused several Chinese AI firms of mounting an “industrial‑scale” distillation campaign against Claude; it has also previously been ordered by a court to pay $1.5 billion over alleged unauthorized use of books. Meanwhile, US AI firms including OpenAI, xAI and Google have likewise traded public accusations over distillation practices. Is this another technical glitch — or evidence of uneven enforcement and double standards in a global race over frontier models? The episode will sharpen scrutiny of both model supply chains and cross‑border AI governance.