TENCENT(HK:00700) Guan Cha Zhe Wang·2025-08-04 11:23Core Insights - Tencent Hunyuan has announced the open-source release of four small-sized models with parameters of 0.5B, 1.8B, 4B, and 7B, designed to run on consumer-grade GPUs and suitable for low-power scenarios such as laptops, smartphones, smart cockpits, and smart homes [1][2] Model Specifications - The models are characterized by low power consumption and high efficiency, with Hunyuan-4B being optimized for smart cockpits and Hunyuan-7B being easily operable on home computers [2] - Hunyuan-4B supports a maximum input of 32K and a maximum output of 32K, while Hunyuan-7B has a maximum input of 16K and a maximum output of 32K [2] - Both models are capable of real-time response, with performance and accuracy being prioritized [2][3] Performance and Compatibility - The new models have achieved leading scores in language understanding, mathematics, and reasoning during testing [3] - They are compatible with mainstream inference frameworks such as SGLang, vLLM, and TensorRT-LLM [8] Unique Features - The models exhibit dual-brain collaboration capabilities, with a "fast brain" for quick responses to simple queries and a "slow brain" for complex tasks, functioning as an efficient assistant [9] - They possess strong memory capabilities, able to handle a context of 256K, retaining details even after multiple discussions [9] - The models also feature advanced agent capabilities, capable of deep information searches, organizing data, and comprehensive travel planning [9]