clauding.de
DE
• 2 min read AI-generated

Kimi K3: The Weights Drop July 27 — and the Theft Accusation Wobbles

Copy article as Markdown

Moonshot releases the full weights of Kimi K3 tomorrow — at 2.8 trillion parameters, the largest open model ever. And right on cue, experts are pushing back on the White House claim that it was distilled from Anthropic's Fable.

Open Source China Anthropic
Featured image for "Kimi K3: The Weights Drop July 27 — and the Theft Accusation Wobbles"

Tomorrow it gets real. On July 27, Chinese lab Moonshot opens up the full weights of Kimi K3 — turning a 2.8-trillion-parameter model into the largest freely available language model ever released. For context: that’s roughly 75 percent bigger than DeepSeek’s V4. And the timing is no accident — it lands right in the middle of the World Artificial Intelligence Conference in Shanghai.

Why this is more than another number

Kimi K3 is a mixture-of-experts model with native vision. On the independent Artificial Analysis leaderboard, it already edges past Anthropic’s Opus 4.8. Once the weights are out, anyone can download the model, self-host it, and build on top. For the open-weight movement, that’s a milestone — and an uncomfortable contrast for the Western labs that keep locking their frontier models down tighter and tighter.

The accusation — and the pushback

A few days ago, the White House’s Michael Kratsios publicly claimed Moonshot had “large-scale” distilled Anthropic’s Fable model to build Kimi K3. A serious charge — theft of intellectual property, essentially.

Now experts are stepping in to question it. Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI, put it plainly to TechCrunch: “You don’t get a model this strong this quickly on the heels of Fable doing strictly distillation. Fable’s only been public since July 1st. You can’t distill that much data, train a model, and release it in two weeks.”

In other words: the timeline barely adds up. If distillation played a role at all, it was marginal at best.

My take

It would be comforting to believe China is only catching up because it copies. That story is soothing — and probably wrong. The less comfortable truth: Moonshot, Alibaba and the rest are now building genuinely good models, with their own architectural ideas like Kimi Delta Attention. Anyone leaning on the distillation accusation will wake up tomorrow morning to find the open weights already running on every server on the planet. And at that point it won’t matter who copied whom — only who does the most with it.

Sources: