OpenAI has introduced new features in its API, including deep research models with support for web search and remote multi-context processing (MCP) for accessing non-public data. These capabilities are available in the o3, o3-pro, and o4-mini models, with web search functionality priced at $10 per 1,000 queries. The deep research models enable developers to build AI agents capable of complex research tasks, code execution, and integration with MCP servers using minimal code. Concurrently, South Korea's LG Group has launched EXAONE 4.0, its flagship hybrid reasoning AI model. The 32-billion parameter EXAONE 4.0 offers toggleable reasoning modes, supports a 131,000-token context window, and holds a non-commercial license. It reportedly outperforms larger models such as DeepSeek R1 and Qwen 235B in instruction following and coding tasks, leveraging architectural innovations like AGAPO advantage, SWA with global attention, and QK-Reorder Layer Normalization to optimize deep layer utilization. EXAONE 4.0 marks South Korea’s first hybrid AI system and is positioned as a significant advancement in AI research and application.