Step 5 Preview gets one-week free trial on Nous Portal
Nous Research launches a 7-day free trial for Step 5 Preview. The MoE model features 600B total parameters, active 27B, and a 1M context window.
Models
Model cards, releases, evaluations, and the limits a lab states in public.
Nvidia Rubin with vLLM: 3.2x More Profit per Gigawatt
SemiAnalysis models Nvidia Rubin running vLLM with 10x performance per dollar and 3.2x profit per gigawatt over GB300 in agentic inference.
H100 Rental Prices Jump 40% to Back $7T AI Debt Market
Power delays pushed Nvidia H100 rental prices up 40% in 2026, creating crucial collateral for a projected $7 trillion AI datacenter debt pipeline.
Microsoft Decision-1 model picks choices without text
Microsoft Decision-1 scores structured options in a single pass without generating text, pricing input at $0.042 per million tokens with free outputs.
Why Chinese Tech Builds In-House AI Models
Chinese giants like miHoYo, Xiaomi, and Meituan fund in-house AI models instead of buying rivals, while distillation blunts tech leads.
Ex-OpenAI researcher cites RLHF limits in AI automation
InstructGPT co-author Diogo Almeida warns that RLHF models suit human assistance, not unsupervised enterprise automation. Here is why current tools fail.
Domyn CEO argues enterprises must own AI models
Domyn CEO Uljan Sharka argues regulated firms must own AI models and data instead of renting commercial APIs from major cloud providers.
OpenAI Launches Ultrafast Mode for GPT-6.1 Sol
OpenAI launches Ultrafast mode for GPT-6.1 Sol across API, Codex, and ChatGPT Work, charging 6x standard rates for up to eight times faster inference.
Zyphra Trains Reasoning Model on Full AMD MI300X Stack
Zyphra completed pre-training for ZAYA1-8B on an AMD MI300X cluster in IBM Cloud, running without Nvidia hardware or software for up to 72 hours.
SemanTok: Stability tests smaller video world models
Stability AI and KIT introduce SemanTok, using semantic predictability to build compact video world models that outperform much larger baselines.