On Poe, a Quora-owned platform that allows users to interact with different AI models, the share of messages to DeepSeek-R1 declined to 3 per cent by the end of April, from a peak of 7 per cent in February, as other affordable reasoning models became available, Poe said in a report published on Wednesday.
These versions of Gemini and Claude fielded 31.5 per cent and 19.1 per cent of Poe subscriber reasoning model queries, respectively, while DeepSeek-R1 was used for 12.2 per cent.
DeepSeek’s foundational V3 model was not among the top five most-used large language models on the platform, according to Poe.
The figures highlight the challenges DeepSeek faces in international markets, despite its breakout success earlier this year. The Hangzhou-based AI start-up became a global sensation in late January with the release of R1, as attention turned to the resource efficiency with which it produced top-performing models.