OpenAI Enhances Codex With Desktop Control and Launches GPT-Rosalind for Drug Discovery

OpenAI has unveiled a dramatically enhanced version of its Codex coding tool, positioning it as a direct competitor to Anthropic's Claude Code with new capabilities that allow it to control desktop applications, generate images, browse webpages, and integrate deeply with users' computers. According to TechCrunch and VentureBeat reports, the updated Codex desktop app for Mac and Windows now operates in the background, accessing other apps on a developer's machine to execute tasks autonomously, marking a shift toward more "agentic" AI that blurs the line between coding assistant and full system operator. This rollout comes as OpenAI confirms reaching 3 million weekly developers, underscoring its growing dominance in developer tools.

The upgrade equips Codex with workplace plugins, persistent memory for past tasks, and the ability to preview web content or create visuals on demand, as detailed in OpenAI's blog post and covered by The Verge via Slashdot. Previously focused on writing, editing, debugging, and shipping code via language models, Codex now aims to evolve into a "Super App" that handles a broader range of workflows. Developers stand to benefit from streamlined productivity, but the expanded permissions raise privacy and security concerns, as the tool gains unprecedented access to local files and apps. OpenAI executives, including CFO Sarah Friar, are increasingly relying on similar AI for mundane office tasks like email summaries, signaling a company-wide pivot toward business users amid competitive pressure from rivals like Anthropic.

In parallel, OpenAI launched GPT-Rosalind, a specialized large language model tuned for life sciences and drug discovery, trained on 50 common biological workflows and integrated with major public databases. As reported by VentureBeat, Ars Technica, and additional announcements, the closed-access model infers protein structures, prioritizes drug targets, maps genotypes to phenotypes, and suggests research pathways while incorporating skeptical tuning to minimize hallucinations and overconfidence. Life sciences researchers, who often navigate fragmented tools and decades of genomic data, could see accelerated discoveries—potentially shortening the typical 10-15 year drug development timeline that costs billions. Access is limited initially to U.S.-based enterprises and partners, with plugins rolling out gradually.

This dual launch reflects OpenAI's strategic refocus on high-value enterprise and scientific applications, as Fast Company notes, even as it faces external hurdles like pausing its massive Stargate data center project in the UK due to energy costs and regulations—a move criticized by Britain's AI minister in Bloomberg coverage. Developers, biotech firms, and businesses are the primary beneficiaries, gaining tools that automate complex tasks and foster innovation. Next steps include broader Codex availability, controlled pilots for GPT-Rosalind with benchmarks for reproducibility, and potential regulatory scrutiny over data handling and AI autonomy. As competition intensifies with Anthropic and others, these updates could redefine how AI integrates into daily work and scientific breakthroughs.