The latest version of the R1 reasoning AI model launched last week by artificial intelligence firm DeepSeek performed well on a number of math and coding benchmarks.
DeepSeek did not disclose the source of the data used to train the model, but some AI researchers speculate that at least some of the data used came from Google's Gemini AI model.
According to a TechCrunch release on Tuesday (3/6), Australian developer Sam Paech, who developed an "emotional intelligence" evaluation for AI, claims to have evidence that DeepSeek's latest model, named R1-0528, was trained using Gemini output.
In a post on the X platform, Paech said that the DeepSeek R1-0528 model's word choice and response style were similar to Google Gemini 2.5 Pro's Gemini output.
That’s not hard evidence. However, another developer who calls himself the creator of an AI free speech evaluation called SpeechMap also noted that R1-0528’s “thought” trace reads like Gemini’s.
