Tuesday, March 25, 2025

Start of archive·May 16

Top story

The Agent RaceGoogle DeepMind Blog

Gemini 2.5: Our most intelligent AI model

Google's latest Gemini iteration introduces reasoning capabilities, signaling an escalation in the model capability race. This is a direct competitive move against OpenAI's o1 and Anthropic's extended thinking models.

Gemini 2.5 released as 'most intelligent' model iteration

The briefs

Enterprise AI agents are moving from capability demos to production deployment at scale. This signals real business impact in high-stakes, regulated industries where automation was previously thought impossible.

Hebbia automation covers 90% of finance and legal work

OpenAI is collapsing the multimodal stack—text, vision, and now generation—into a single model interface. This changes how builders integrate image workflows and signals where the industry is heading on unified model architecture.

Image generation now integrated directly into GPT-4o (not a separate product)

OpenAI is expanding GPT-4o's multimodal reach beyond text and vision-understanding into generative image creation, directly challenging specialized image models and expanding the moat of its flagship model.

GPT-4o image generation described as 'significantly more capable' than DALL·E 3

Vercel's AI Gateway Custom Reporting lets founders track and optimize AI spend across models and providers in one place—critical as inference costs become a line-item business decision for startups and enterprises scaling LLM apps.

Custom Reporting API now in beta for Pro and Enterprise users