OpenAI and partners roll out enterprise models

OpenAI spent this week pushing its business AI stack deeper into real-world work, with new models for labs, security teams, and enterprise software builders. (openai.com 1) (openai.com 2) The company updated its Agents software development kit on April 15 with sandbox agents in the Python SDK, giving developers container-based environments with files, commands, packages, ports, snapshots, and memory for multi-step tasks. OpenAI’s developer docs say the toolkit is for applications that plan, call tools, and keep state across longer jobs. (developers.openai.com) (techcrunch.com) A sandbox is a locked room for code: the agent can test software and handle files inside that room without getting broad access to the rest of a company’s systems. TechCrunch reported the update also added a frontier-model harness so companies can deploy and test longer-running agents with approved tools in controlled workspaces. (techcrunch.com) (developers.openai.com) On April 16, OpenAI introduced GPT-Rosalind, a research-preview model for biology, drug discovery, and translational medicine. OpenAI said the model is available in ChatGPT, Codex, and the application programming interface for qualified customers through its trusted access program. (openai.com) (reuters.com) OpenAI said GPT-Rosalind is tuned for chemistry, protein engineering, and genomics, and can help with literature review, sequence-to-function interpretation, experimental planning, and data analysis. The company said drug development in the United States typically takes roughly 10 to 15 years from target discovery to regulatory approval. (openai.com) OpenAI is pitching the model at customers including Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific. Reuters reported demand for AI tools in drug discovery has risen across pharmaceutical companies, academic institutions, and biotech firms. (openai.com) (reuters.com) The cybersecurity release is narrower. On April 14, OpenAI unveiled GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for defensive work and distributed through its Trusted Access for Cyber program rather than a broad public launch. (openai.com) (reuters.com) OpenAI said it is expanding that cyber program to thousands of verified defenders and hundreds of teams protecting critical software. The company also said Bank of America, BlackRock, Cisco, Cloudflare, CrowdStrike, Oracle, Palo Alto Networks, and Zscaler are among the organizations already signed up to support the effort. (openai.com 1) (openai.com 2) Cloudflare, meanwhile, said on April 13 that it was expanding Agent Cloud so developers can build, deploy, and scale agents across its network. OpenAI said Cloudflare customers can use GPT-5.4 and Codex inside that platform, and Cloudflare said the goal is to move agents from laptop demos to production workloads. (cloudflare.com) (openai.com) OpenAI also said Codex is now on Windows. Its release notes say the Windows app supports multiple Codex agents in parallel with isolated worktrees and reviewable diffs, and OpenAI’s developer docs say it can run natively with PowerShell and a Windows sandbox or through Windows Subsystem for Linux 2. (help.openai.com) (developers.openai.com) Taken together, the releases point to the same target customer: companies that want AI systems to do longer, riskier, and more specialized work inside controlled environments instead of a chatbot window. (developers.openai.com) (openai.com)