AI Pulse: Bigger models, managed agents, and tighter control

OpenAI has introduced GPT-6 Astra as a business-focused model with stronger reasoning, computer-use abilities, and improved writing and design judgment. For teams building AI tools, the interesting shift is toward models expected to handle broader work tasks—not just answer questions—though evaluation in real workflows will matter more than launch claims.

OpenAI’s new Agents API packages orchestration, context handling, and execution infrastructure into one service. That could reduce the amount of plumbing teams need to build themselves, but it also raises practical questions about portability, observability, and how much control developers retain over agent behavior.

OpenAI has launched a version of ChatGPT designed for financial-services professionals, combining its latest model with built-in data from providers. The reported focus on research, modeling, and pitchbook work shows how AI vendors are targeting specific job functions—and puts a premium on accuracy, auditability, and responsible use in high-stakes settings.

Sakana AI has released Fugu Max and Fugu Ultra v2, presenting them as cheaper and stronger options for multi-agent orchestration. The reminder that Fugu is a family rather than one foundation model is useful: builders may increasingly choose specialized combinations of models and coordination layers instead of relying on a single general-purpose system.

My AI Fact Book has launched a platform focused on tracking AI models, pricing, and industry changes, with an emphasis on verification. For practitioners comparing vendors or planning budgets, dependable historical information could become as valuable as benchmark scores—especially as model names, prices, and capabilities keep changing.

F5 has introduced Workforce AI Security to monitor employee use of AI tools and the actions taken by AI systems. This reflects a growing enterprise concern: organizations are not only deciding which models to approve, but also trying to understand what people and automated systems actually do with them.

OPAQUE has launched a Weight Custody Manifest standard aimed at sovereign and on-premises AI environments. If adopted, a verifiable way to document where model weights are held and controlled could help organizations that need more confidence over deployment location, ownership, and operational custody.