Tencent Open-Sources 770B-Parameter Hy4 AI Model; OpenAI's Jalapeño Accelerator Shows Strong Performance

Tencent has open-sourced a preview of its Hy4 AI model under the Apache 2.0 license. This large model features 770 billion total parameters (49 billion active) and a context window exceeding 1 million tokens, targeting applications in coding, office work, and scientific research. In a separate development, OpenAI published inference results for its Broadcom-built Jalapeño accelerator, claiming 1.5–1.9x more work per watt and 1.7–3.6x lower end-to-end latency than Nvidia Blackwell systems across various open models. These advancements highlight significant progress in both the accessibility of large-scale AI models and the efficiency of AI compute infrastructure.