OpenAI is spending heavily on regulatory alignment and institutional relationships while shipping product improvements that matter to paying customers. The $5 million commitment to Lenfest, the safety cases framework, and the Australia apology represent a clear strategy: establish OpenAI as the responsible operator willing to work within government structures, preempt criticism before it hardens into policy, and lock in media and civic institutions as stakeholders in the company's success. Meanwhile, GPT-6 Astra's 2x speed improvement on tax workbooks shows where the real value accrual happens, not in benchmarks or safety papers, but in concrete workflows that reduce friction and cost for enterprise users willing to pay. NVIDIA's $150 billion repurchase authorization signals confidence in sustained demand for AI compute infrastructure, but the Open Agent Safety Platform announcement is the tell: NVIDIA is positioning itself as the neutral infrastructure layer that can sell tools to every player regardless of model choice, turning safety and governance into a hardware-plus-software bundle that locks in NVIDIA's position across the stack. AWS, AMD, and Hugging Face are all moving toward agent-based computer use and reinforcement learning infrastructure, each betting that the next layer of value sits in orchestration and control rather than base model capability. The collective pattern is not competition on intelligence alone but consolidation around who owns the deployment layer, the safety narrative, and the relationships with institutions that will regulate this space.
Sloane Duvall
A curated reference of models from major AI labs, with open/closed weight status, input modalities, and context window size. American labs tend towards closed weights models and Chinese labs tend toward open weights models.
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None