Guan Cha Zhe Wang·2025-05-14 04:08Core Insights - The usage of the DeepSeek-R1 model by the Chinese company DeepSeek has decreased by 50% from its peak in February, yet it remains in third place among inference models [1][3] - Kuaishou's Kling series of video generation models has rapidly gained over 30% market share, with Kling-2.0-Master achieving 20.9% within three weeks of its release [1][5] Inference Model Trends - The "DeepSeek moment" in February caused the share of inference models in all text models to surge from 2% to 10% within two weeks, currently stabilizing at 8% [1][3] - DeepSeek-R1 captured over 50% of the inference model text messages sent to the platform shortly after its launch, breaking OpenAI's previous monopoly [3] - As of March, the entry of Anthropic's Claude-3.7-Sonnet-Reasoning model led to a decline in DeepSeek-R1's market share, which was further impacted by Google's Gemini-2.5-Pro, now holding 31.5% [3][5] OpenAI and Competitors - OpenAI's inference model family has maintained a total market share of no less than 30% due to continuous releases of various models [5] - Grok 3 model has less than 1% market share, possibly due to limited API support for its mini version [5] Video Generation Models - Kuaishou's Kling series has a combined market share exceeding 30%, with Runway leading individual model shares at 23.6% [5] - Kling-2.0-Master supports high-definition video generation at 1080p and has seen rapid adoption, reaching a user base of over 22 million since its launch [7]