Moonshot AI and DeepSeek reportedly used OpenAI’s GPT-6 Astra outputs to build new models

The Chinese labs released upgraded models this month with reportedly significant help from an unexpected source. Moonshot AI unveiled Kimi K3.1 Pro on October 23, 2026, while DeepSeek launched DeepSeek-V4.1-Flash-1017 on October 17, 2026. Both models reportedly incorporated outputs from OpenAI's GPT-6 Astra, sparking debate about whether this constitutes clever engineering or improper copying.

Model Details

DeepSeek's latest model focuses on improved task performance and enhanced agentic coding capabilities, allowing it to plan and execute multi-step programming work with minimal human intervention. The company maintained its signature open-weight, low-cost approach that first brought attention to the lab.

Moonshot's Kimi K3.1 Pro features approximately 2.8 trillion parameters and scored 55 on the Artificial Analysis Intelligence Index, ranking fourth overall. Notably, it outperformed GPT-6 Astra on certain agentic benchmarks including Terminal-Bench 4.0, demonstrating that a model reportedly built with Astra's help can now surpass the original on specific tests.

Distillation Concerns

The practice at issue is known as distillation, where a model learns from another's outputs without direct access to its training data. This technique allows labs to absorb much of a teacher model's capability at a fraction of the training cost. Both Chinese labs have faced similar allegations before, with Anthropic accusing them in February 2026 of running distillation attacks on its Claude models using fraudulent accounts.

Financial Implications

These developments carry significant financial weight. Moonshot AI's valuation reportedly reached $50 billion by October 2026, ahead of a planned Hong Kong IPO. DeepSeek has achieved even higher valuations in recent funding rounds. The performance numbers support their competitive pitch, but distillation reports complicate their narratives about technological independence.

For Moonshot, the immediate stakes involve its upcoming IPO, where investors will scrutinize training practices. For DeepSeek, its low-cost, open-weight approach could face increased scrutiny if questions about the origin of its training signal persist. The industry will be watching closely for OpenAI's response, as formal policy debates about model ownership may emerge if the company follows Anthropic's lead in making public accusations.