Hunyuan-A13B-Instruct activates only 13 B of its 80 B parameters, yet matches much larger LLMs on mainstream benchmarks. It offers hybrid reasoning: low-latency “fast” mode or high-precision “slow” mode, switchable per call. Native 256 K-token context lets it digest book-length documents without degradation. Agent skills are tuned for BFCL-v3, τ-Bench and C3-Bench leadership, making it an excellent autonomous assistant backbone. Grouped Query Attention plus multi-format quantization delivers memory-light, GPU-efficient inference for real-world deployment, with built-in multilingual support and robust safety alignment for enterprise-grade applications....

Hy4 preview is Tencent Hy's new-generation productivity flagship model, built on Gated DeepSeek Sparse Attention with iHC residual design, with approximately 770B total and 49B activated parameters per token. It natively supports a 1M context window with up to 64k output and a native MTP layer for speculative decoding to boost generation throughput. It demonstrates strong comprehension, planning, and sustained execution on complex tasks, and natively supports tool calls, structured output, and context caching. Deeply optimized for Agent, Coding, and productivity scenarios, it suits long-horizon planning and code workflows....

Built for real-world business scenarios, Hy3 features a 295B/21B active MoE architecture, native 256K context support, and three reasoning modes. It enhances coding, long-form comprehension, multi-turn dialogue, and agentic task execution, balancing reliability, efficiency, and cost across both high-frequency interactions and complex workflows....