Why Dwarkesh is Wrong about Computer Use + How OpenAI shipped its Jev competitor in 1 Week
Latent Space's OpenAI DevDay podcast covers GPT-6.1 Sol, Dots, Computer Use agents, and the new Agents API and Decisions API developer stack.
Ari Weinstein, who leads OpenAI's Computer Use product and engineering, explains how agents now combine screenshots, accessibility data, the DOM, Playwright and generated code, moving from human-level toward superhuman software use. Nikunj Handa details the new developer stack: async tool calling, mid-turn steering, WebSockets, UltraFast inference, the Decisions API, prompt caching, pre-warming, compaction, and the Agents API. GPT-6.1 Sol is cited as a fifth of the cost of Astra and a seventh for Computer Use workloads. The Decisions API is described as initially a Luna wrapper built in a roughly one-week sprint.
- GPT-6.1 Sol costs a fifth of Astra and a seventh for Computer Use workloads
- Agents API opens the Computer Use behind Codex and ChatGPT to developers
- Decisions API shipped in about one week, initially as a Luna wrapper
- Computer Use agents now combine screenshots, accessibility trees, DOM, Playwright and generated code
- Organizations
- OpenAIAnthropicAri WeinsteinNikunj Handa
From the original article
15 min read at latent.spaceOur DevDay coverage - the first pod on the DevDay lineup - dives in with the leaders of OpenAI’s CUA team and API platform.
latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.