Search results for CAISI
AI & Enterprise
DeepSeek keeps prices 75% lower with GPT-level performance; memory efficiency seen as key
China\'s AI company DeepSeek is in talks to raise about 70 billion yuan while maintaining free model releases and a low-cost API strategy, with efficiency in computing and memory seen as its core edge. Analysts cited sharply lower memory use versus rivals and optimised cache design that cuts costs and improves inference speed even on lower-performance chips. The approach is also linked to constraints on access to Nvidia\'s newest AI chips.
AI & Enterprise
DeepSeek V4 Pro nears GPT-5-level performance, rated best Chinese AI model
DeepSeek’s latest model, DeepSeek V4 Pro, is about 8 months behind top U.S. AI models, a U.S. government-affiliated assessment showed. The AI Standards and Innovation Center under NIST said it delivered the highest performance among Chinese AI models, scoring about 200 points above Kimi K2.5. CAISI also rated it more cost-efficient than comparable models, outperforming OpenAI’s GPT-5.4 Mini on 5 of 7 benchmarks.
AI & Enterprise
South Korea\'s ICT ministry eyes access to Anthropic\'s Mythos
South Korea\'s Ministry of Science and ICT is actively exploring participation in Anthropic\'s cybersecurity consultative body, Project Glasswing. The ministry held a meeting with Anthropic and multiple Korean agencies on AI and cybersecurity cooperation and conveyed broad interest in collaboration, including information sharing on vulnerability disclosures. Prospects remain uncertain due to reported White House opposition to expanding the project and the absence of an MOU with Anthropic. The ministry is also accelerating domestic responses to AI-driven cyber threats.