OpenAI’s New AI Chip Just Got Real (Beats NVIDIA)

OpenAI’s Jalapeño AI chip just posted some wild new results against NVIDIA, including huge efficiency gains at high decoding speeds. But that’s only part of the story, as Anthropic quietly tests mystery Claude models, Alibaba makes Qwen dramatically cheaper, and Claude gets a major memory upgrade. 📩 Brand Deals & Partnerships: [email protected] ✉ General Inquiries: [email protected] What You’ll See: OpenAI’s Jalapeño inference chip beat NVIDIA’s GB200 and GB300 systems on efficiency and latency across several major AI models, with some extreme workloads producing much larger gaps. SOURCE: https://www.theverge.com/ai-artificial-intelligence/984290/openai-jalapeno-ai-chip-benchmarks Anthropic’s mysterious Melon and Marshmallow models briefly surfaced in developer communities before disappearing, fueling speculation around Anthropic’s next Claude models and a possible Fable 5.1 rollout. Anthropic has not officially confirmed that either codename is Fable 5.1. SOURCE: https://www.orcarouter.ai/blog/claude-marshmallow-melon-eap-leak Alibaba launched Qwen3.8-Flash with stronger coding and office capabilities while dramatically cutting training costs compared with Qwen3.7-Plus. SOURCE: https://www.reuters.com/business/retail-consumer/alibabas-qwen-launches-qwen38-flash-ai-model-with-lower-training-costs-2026-08-26/ Claude now shares memory between regular chat and Cowork, allowing project context to carry across both instead of forcing users to explain everything again. SOURCE: https://techcrunch.com/2026/08/25/claude-cowork-finally-remembers-what-you-told-the-app-in-chat/ 🚨 Why It Matters OpenAI is turning custom silicon into a serious part of its AI infrastructure while NVIDIA faces growing pressure from its own biggest customers. At the same time, Anthropic and Alibaba are pushing faster model cycles, cheaper AI, and agents that remember more of your work. #ai #openai #nvidia