Alibaba released its most powerful AI model yet on August 2, 2026, a 2.4-trillion-parameter model called Qwen3.8-Max. It landed in the middle of an escalating, unresolved dispute: Anthropic has accused Alibaba’s Qwen lab of running the largest AI model-copying campaign it has ever documented, using tens of thousands of fake accounts to extract Claude’s capabilities.
Quick facts
- Qwen3.8-Max has 2.4 trillion parameters, ranking fifth in Text Arena and second in Vision Arena, with open weights planned for release the following week.
- In a June 10, 2026 letter to the US Senate Banking Committee, Anthropic alleged operators tied to Alibaba’s Qwen lab used roughly 25,000 fraudulent accounts to generate 28.8 million exchanges with Claude between April 22 and June 5.
- Anthropic says the campaign specifically targeted Claude’s software engineering, agentic reasoning, and cybersecurity capabilities, tied to its Mythos model line.
- Alibaba disputes the allegations; no regulatory action or independent verification of the claims has been confirmed as of this writing.
- In a separate twist, developers have reported Claude Opus 4.8 identifying itself as “Qwen” during some Chinese-language tests, which critics have pointed to as ironic given Anthropic’s own accusations.
What Qwen3.8-Max actually is
Per AI Magazine’s reporting, Alibaba CEO Eddie Wu’s team is positioning Qwen3.8-Max as the company’s most capable model to date, sitting just below Moonshot’s 2.8-trillion-parameter Kimi K3 in raw size but ranking competitively on public leaderboards. It’s live now on Alibaba Cloud’s Model Studio APIs and through QwenWork, with open weights due out the following week, continuing Alibaba’s return to open-sourcing its flagship models after several proprietary-only releases earlier in the year. Notably, Alibaba hasn’t published token pricing yet, though its recent models have consistently undercut Western rivals on cost.
The distillation accusation behind the release
Model distillation is a real, well-understood technique: a smaller or newer model is trained by repeatedly querying a more capable model and learning from its outputs, letting the newer model absorb capability without bearing the original training cost. Anthropic’s letter to the Senate Banking Committee, reported by multiple outlets, describes the alleged Alibaba campaign as roughly 1.7 times larger than three prior Chinese distillation campaigns Anthropic disclosed combined (from DeepSeek, Moonshot AI, and MiniMax, totaling about 16.5 million interactions in February 2026). It’s worth being precise about the epistemic status here: the 28.8 million figure and 25,000 account count are Anthropic’s own allegations, made in a letter to Congress rather than an independently audited report, and Alibaba has denied wrongdoing.
Why the “Claude calls itself Qwen” detail complicates the narrative
Developers testing Claude Opus 4.8 in Chinese-language conversations have reported the model occasionally self-identifying as “Qwen” rather than Claude, which critics on social media have seized on as evidence of hypocrisy given Anthropic’s own distillation accusations against Alibaba. It’s worth some caution here too: models misidentifying themselves is a known, previously documented phenomenon across the industry, often traced to training data that includes text about other AI systems, rather than proof of one company training directly on a competitor’s outputs. The detail is genuinely newsworthy as a complicating wrinkle in the public narrative, not as confirmation of anything about how either model was actually built.
The bigger picture: an escalating US-China AI dispute
This lands inside a broader pattern of accusation and countermeasure. Days after Anthropic’s June letter, the US Commerce Department restricted Anthropic’s own Fable 5 and Mythos 5 models on national-security grounds; the Pentagon separately added Alibaba to a restricted list around the same period; and Alibaba has reportedly pushed its own staff toward its in-house Qoder coding platform instead of Claude Code, the same platform now hosting the Qwen3.8-Max preview. Whatever the truth of the distillation claims, both companies are clearly treating model capability, and control over how it was obtained, as a matter of active competitive and national-security consequence, not just an engineering question.
Key takeaway
Qwen3.8-Max is a genuinely capable model shipping into a market that will judge it as much on the distillation dispute swirling around it as on its benchmark scores. Treat the specific numbers on both sides, Anthropic’s 28.8 million query claim and Alibaba’s leaderboard rankings, as allegations and marketing respectively until independent parties verify either one.
Leave a Reply