Alibaba's Qwen3.8-Max is multimodal (text, images, video) with a 1M-token context and outperforms Claude Fable 5 and GPT-5.6 Sol across 7 coding and 36 multimodal benchmarks. Verified scores include GPQA Diamond 92.6%, SWE-bench Pro 67.7%, Terminal-Bench 2.1 86.6%, and OSWorld-Verified 86.1%; the model also completed a 16-day software engineering project autonomously.
The Trump administration finalized a voluntary AI oversight framework but will not publicly release it, and the classified benchmarking process explicitly excludes open-weight models. Chinese open-weight models will be spared US tests, while OpenAI, Anthropic, Google, Meta, Nvidia, and Microsoft attended a closed briefing on the framework.
Andrew Ng and other US AI leaders are now using Chinese open-weight models including Moonshot's Kimi K3 and Zhipu's GLM-5.2 for security reviews, challenging closed-source safety claims. DeepSeek's V4-Flash is pushing prices down, and The Register reports US model makers are panicking as China floods the market with open releases.
OpenAI reduced GPT-5.6 Luna API prices by 80% and Terra by 20%, citing GPU kernel improvements and better speculative decoding. Box CEO Aaron Levie warns AI pricing could collapse as open-weight models like Qwen close the capability gap, potentially triggering a race to the bottom that squeezes closed-source rivals.
Google has built a $150 billion infrastructure financing program to supply Anthropic with AI chips, involving Broadcom, Apollo, and Blackstone, covering everything from chip fabrication to data center deployment. Separately, AI cloud startup Volta was valued at $2.4bn and inked a $10bn deal with Anthropic, which recently also partnered with AMD for 2GW of compute.
AI server shipment growth for 2026 was revised up to 31% year-over-year, driven by continued investment in GPU clusters and liquid-cooling infrastructure. Amazon, Google, Meta, and Microsoft have collectively spent more than $1 trillion on AI infrastructure since 2023, with an additional $745 billion expected to be added in 2026 alone.