Full Breakdown
JoyIn Accuses OpenAI of Direct Model Distillation
By Drooid · · How we work
Core Event
In a publicly released Chinese-language letter, Gu Guo Renjie, chief executive of Suzhou-based JoyIn—a humanoid robotics startup backed by Ant Group—asserted that OpenAI’s recently launched GPT-6 Astra model is a “direct distillation” of JoyIn’s Aether model with no modifications. The claim focuses on parallel technical approaches such as “recursive self-improvement” and AI-driven computing-power optimization, as well as a shared space-themed visual design on the two companies’ web pages. JoyIn announced that it has begun legal proceedings over the alleged copying.
Background & Context
The dispute unfolds amid an intensifying U.S.–China competition in artificial-intelligence research. U.S. firms, notably Anthropic, have repeatedly accused Chinese companies of illicitly distilling their models to boost performance. A U.S. cybersecurity agency recently reported that six Chinese firms—including DeepSeek and Alibaba—had distilled models from Anthropic, Google and OpenAI. JoyIn’s allegation reverses the usual direction, positioning OpenAI as the accused party for the first time in this “distillation” debate.
Timeline
- Weeks before September 6 – JoyIn demonstrated its “extraterrestrial visitor” AI framework at a Silicon Valley event.
- September 6 – OpenAI chief scientist Jakub Pachocki published an article titled “An Alien Mind.”
- Early September – OpenAI released GPT-6 Astra.
- Early September – JoyIn publicly announced the Aether model and filed a lawsuit alleging direct distillation.
Data & Statistics
- JoyIn’s Aether model claims a 90 % success rate for humanoid robots completing tasks on the first attempt.
- Development lead Zhu Mingxuan said Aether reduces training time by two-thirds compared with prior approaches.
- JoyIn plans to open-source tactile-sensing modules of Aether while keeping energy-usage components proprietary.
Official Statements & Responses
OpenAI did not respond to requests for comment.
Garry Tan, chief executive of startup accelerator Y Combinator, said he would “do nothing” about the distillation claim and noted that U.S. companies have trained models on data protected by copyright law.
The U.S. cybersecurity agency’s recent disclosure about other Chinese firms’ distillation activities underscores the broader regulatory scrutiny of cross-border AI model reuse.
Verbatim Quotes
- “People often say major tech companies have intelligence networks monitoring the whole internet, this time I believe it, this is a direct distillation of us without any modifications,” — Guo Renjie, joyin's CEO
Conflicting Reports & Gaps
- CNBC and other outlets were unable to independently verify JoyIn’s technical similarity claims.
- The allegation that OpenAI accessed JoyIn’s publicly released content “directly” remains unsubstantiated pending the outcome of the lawsuit.
What’s Next
JoyIn’s legal filing initiates a formal dispute that could set precedent for how “distillation” is interpreted across jurisdictions. Observers anticipate further statements from OpenAI and possible regulatory review as the case proceeds.
