Modelos disponibles

DeepSeek V4-Flash

Por defectodeepseek-v4-flash

el más rápido

GLM-5.3

glm-5.3

GLM-5.3 is Zhipu's latest flagship model, with comprehensive advancements in complex software engineering and agentic task capabilities. It shares the same base model as GLM-5.2—all improvements come from post-training. Compared to GLM-5.2, it demonstrates superior performance in complex programming and long-horizon tasks: -Stronger programming capabilities GLM-5.3 delivers a significant boost in programming ability, achieving a 50% improvement over GLM-5.2 on Zhipu's internal Z.ai Code Bench, and reaching open-source SOTA performance on public benchmarks including Terminal Bench 3.0 and Agents' Last Exam (CLI). -Emergent cybersecurity capabilities As post-training scale continues to expand, the model's cybersecurity capabilities improve at a faster-than-expected pace. GLM-5.3 achieves state-of-the-art results on the vulnerability discovery benchmark CyberGym; the deeper it goes into the exploit chain, the more pronounced its gains over GLM-5.2 become, with scores more than doubling those of GLM-5.2 on exploit-oriented benchmark tests.

DeepSeek V4.1-Flash

deepseek-v4.1-flash

DeepSeek-V4.1-Flash is a new-generation large language model officially released by DeepSeek on September 10, 2026. The model uses a 552B-parameter MoE (Mixture-of-Experts) architecture, paired with an all-new asymmetric Causal-Encoder-Decoder structure. During the prefill stage, it activates only about 8B parameters per token, and during the decode stage, it activates about 16B parameters. In terms of KV cache compression, V4.1 Flash reduces HBM memory requirements to one-quarter of the previous generation and SSD storage requirements to one-eighth, significantly lowering the usage cost of agent-style long-context tasks.