OpenAI has announced a strategic pivot away from its broad "do everything" product approach, concentrating instead on coding tools and business customers. This shift comes as the company's leadership acknowledges it has been scrambling to keep up with rival Anthropic's momentum in the enterprise AI sector.
The strategic reassessment was communicated to employees by Fidji Simo, OpenAI's CEO of Applications, during a company-wide meeting last week. Simo stated that senior leaders, including CEO Sam Altman and Chief Research Officer Mark Chen, are actively reviewing which products to cut or scale back. "We cannot miss this moment because we are distracted by side quests," Simo told staff, according to remarks reviewed by The Wall Street Journal. "We really have to nail productivity in general and particularly productivity on the business front." A formal update on the strategic changes is expected in the coming weeks.
This refocus follows a year of rapid and diverse product releases from OpenAI, including the video generator Sora, a web browser called Atlas, a hardware device, and e-commerce features for ChatGPT. Altman had previously compared this model to "betting on a series of startups" internally. However, the pace created internal confusion and an unwieldy structure, with computing resources shifting between teams at the last minute. For example, the Sora team sat within the research division despite launching a major consumer product. The Sora standalone app, which launched with a social feature last September, briefly topped Apple's App Store but later lost momentum. OpenAI now plans to fold its video tools into the main ChatGPT app.
Anthropic's success with its Claude Code and Cowork products, which have made it the dominant AI provider for businesses, served as a "wake-up call" for OpenAI. These products, popular with software engineers for handing off complex tasks to AI agents, have even triggered a global stock-market selloff. Unlike OpenAI, Anthropic has avoided image and video generation to maintain a sharp focus on enterprise and coding.
In response, OpenAI has accelerated its efforts in this domain. It released a new version of its Codex app last month and a model called GPT-5.4 tailored for professional use. Simo announced on X that Codex now boasts over two million weekly active users, a nearly fourfold increase since the start of the year. The company is also deploying engineers with consulting firms and business partners to drive AI adoption across industries.
OpenAI is also benefiting from geopolitical developments. The Pentagon recently labeled Anthropic a supply-chain risk after the company refused to allow unrestricted military use of its AI. Anthropic is suing the U.S. government over this designation. Meanwhile, OpenAI has reportedly struck a new agreement with Amazon Web Services to bring its AI tools to federal employees.
Concurrently with its strategic shift, OpenAI launched two new AI models, GPT-5.4 Mini and GPT-5.4 Nano, on March 17, 2026. These are not stripped-down versions but purpose-built models designed for high-volume, latency-sensitive workloads like customer support and automated workflows, where speed and cost are more critical than peak accuracy.
OpenAI calls them its "most capable small models yet," with GPT-5.4 Mini being more than twice as fast as its predecessor. On benchmarks, GPT-5.4 Mini scored 54.4% on the SWE-Bench Pro coding test (vs. 57.7% for the full GPT-5.4) and 72.1% on the OSWorld-Verified test for desktop operation. GPT-5.4 Nano scored 52.4% and 39.0% on the same tests, respectively. The pricing is designed for scale: GPT-5.4 Mini costs $0.75 per million input tokens, while Nano costs just $0.20 per million input tokens.
Financially, both OpenAI and Anthropic are in talks with private equity firms ahead of potential public listings. OpenAI is in advanced discussions with TPG, Advent International, Bain Capital, and Brookfield Asset Management for a joint venture valued at roughly $10 billion pre-investment. The company also faces a new lawsuit from Encyclopedia Britannica and Merriam-Webster, accusing it of scraping protected content to train ChatGPT models.
