# Moonshot AI and DeepSeek reportedly used GPT-6 Astra outputs to build new models

Digest AI · Generative AI & Models · published 2026-10-27T18:10:00Z

Canonical: https://digestai.news/story/moonshot-ai-and-deepseek-reportedly-used-gpt-6-astra-outputs-to-build

## Summary

Moonshot AI and DeepSeek, two of China’s most watched AI labs, each released upgraded open‑weight models in October. DeepSeek launched **DeepSeek‑V4.1‑Flash‑1017** on October 17, 2026, emphasizing sharper task performance and “agentic coding,” where the model plans and executes multi‑step programming without hand‑holding. Moonshot followed on October 23, 2026 with **Kimi K3.1 Pro**, a model of roughly **2.8 trillion parameters** that scored **55** on the Artificial Analysis Intelligence Index, placing it fourth overall and beating OpenAI’s GPT‑6 Astra on the Terminal‑Bench 4.0 agentic benchmark.

Both labs say they built the new models using outputs from OpenAI’s GPT‑6 Astra, a practice known as distillation—feeding a strong model’s responses into a new model’s training. Anthropic previously accused Moonshot and DeepSeek of running similar distillation attacks on its Claude models, alleging use of fraudulent accounts. Moonshot’s valuation is reported at **around $50 billion** ahead of a planned Hong Kong IPO, while DeepSeek has been valued even higher in recent funding rounds. The success of these open‑weight models raises questions about the durability of competitive advantage when training relies on a rival’s outputs.

## Key points

- DeepSeek‑V4.1‑Flash‑1017 launched Oct 17 2026 with agentic coding capabilities.
- Moonshot’s Kimi K3.1 Pro (≈2.8 trillion parameters) scored 55, fourth on AI Index, beating GPT‑6 Astra on Terminal‑Bench 4.0.
- Both labs reportedly used GPT‑6 Astra outputs for training, echoing earlier Anthropic distillation accusations.

## Why it matters

Open‑weight models that match or exceed a flagship OpenAI system show China’s labs can compete at the frontier, while reliance on another model’s outputs raises legal and investment‑risk questions.

## Sources

1. [Moonshot AI and DeepSeek reportedly used OpenAI’s GPT-6 Astra outputs to build new models](https://cryptobriefing.com/moonshot-deepseek-openai-astra-models) (cryptobriefing.com, 2026-10-27)

Part of the developing story: [Moonshot AI's Rapid Rise to Global AI Power](https://digestai.news/thread/moonshot-ai-releases-kimi-k3-1-pro-with-1m-token-context-and-open-weights-by) (4 stories)

## Cite

Digest AI, "Moonshot AI and DeepSeek reportedly used GPT-6 Astra outputs to build new models", 27 October 2026, https://digestai.news/story/moonshot-ai-and-deepseek-reportedly-used-gpt-6-astra-outputs-to-build

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/moonshot-ai-and-deepseek-reportedly-used-gpt-6-astra-outputs-to-build.json