3Before9

Stanford AI Index Confirms China Has Effectively Closed the Gap With the US

Stanford's Human-Centred AI Institute released its 2026 AI Index report this week, and the headline finding is stark: the performance gap between the best American and Chinese AI models has narrowed to just 2.7 per cent. ByteDance's Dola-Seed-2.0 Preview now trails Anthropic's Claude Opus 4.6 by only 39 Elo points on leading benchmarks, and the lead has changed hands multiple times over the past year. China is also dominating in patents, research publications, and what the report calls "physical AI" - autonomous robotics and real-world control systems. Southeast Asian countries including Singapore, Malaysia, Thailand, and Indonesia are trending more positive towards AI adoption, with Singapore's adoption rate hitting 61 per cent, well above the global average.

SoftBank, NEC, Sony and Honda Form Joint Venture to Build Japan's Own AI Foundation Model

Four of Japan's largest technology and industrial companies - SoftBank, NEC, Sony and Honda - have established a new entity called Japan AI Foundation Model Development, with the explicit goal of building a domestically owned physical AI foundation model. The venture is targeting a roughly one-trillion-parameter model tuned for real-world control tasks such as autonomous driving and industrial robotics, trained on Japanese data without routing through foreign cloud platforms. SoftBank and NEC will lead development, Honda plans to deploy the model first in its autonomous vehicles, and Sony will contribute robotics and gaming hardware expertise. The investor list extends well beyond the four founders to include Nippon Steel, Kobe Steel, and all three of Japan's megabanks - MUFG, Sumitomo Mitsui and Mizuho.

Huawei Cloud Launches Model-as-a-Service Token Platform Across Five Asia Pacific Markets

Huawei Cloud used an AI Boost Day event in Jakarta on 15 April to officially launch its Model-as-a-Service offering across Asia Pacific, giving enterprise customers a managed token service for running inference on mainstream large language models. The platform supports six models spanning three families - GLM, DeepSeek and Qwen - optimised for intelligent Q&A and AI coding use cases. Huawei's in-house acceleration engine underpins the service, which runs on an infrastructure network of five regions and 18 availability zones across Singapore, Thailand, Hong Kong, Indonesia and the Philippines, promising sub-50-millisecond access latency.