Moonshot AI and DeepSeek relied heavily on outputs from OpenAI’s GPT-6 Astra to train their latest models, The Information reported, citing people with direct knowledge, according to Reuters.
Reuters reported that The Information identified Moonshot’s Kimi K3.1 Pro, released the week before Reuters’ Oct. 28 report, and DeepSeek’s upgraded V4.1-Flash 1017, launched Oct. 17, as models that benefited from distilling Astra. Distillation, in this account, involves using one model’s outputs to train another.
In a separate matter, OpenAI said in a Sept. 30 blog post that it had linked a core part of a distillation campaign begun in July to individuals associated with Moonshot AI, Reuters reported. That statement does not establish that the campaign involved the reported Astra use.
The report drew differing reactions. kimmonismus called the claim damaging to Kimi’s reputation. agihippo argued in a quote-post that a model trained through distillation cannot qualify as “frontier,” while __tinygrad__ replied that Kimi should distill “Opus 5.5 harder” and disputed the reputation criticism. Those are the posters’ views, not findings about Kimi’s reputation or a settled standard for frontier models.
At publication, Reuters said OpenAI, DeepSeek and Moonshot AI had not immediately responded to its requests for comment.
