SAN FRANCISCO, March 2026 — OpenAI is gearing up to launch successors to its flagship models, including GPT-5 and an upgraded version of GPT-4o, with internal timelines pointing to a major release in the first half of 2026, according to sources familiar with the company’s development roadmap.
The company has been quietly testing next-generation models under code names including “Strawberry” and “Orion,” which are designed to deliver significantly improved reasoning, planning, and multi-step task execution compared to GPT-4o.
These models are expected to power a new wave of AI agents capable of handling complex, long-horizon workflows—such as booking travel, managing finances, or coordinating multi-app tasks—autonomously or with minimal human intervention.
OpenAI CEO Sam Altman has repeatedly signaled that 2026 will be the year AI agents move from experimental to practical everyday use. In recent interviews, Altman described the upcoming models as having “new capabilities in reasoning and tool use that are meaningfully better” than current systems.
He also indicated that GPT-5 (or its equivalent) will represent a substantial leap in intelligence, potentially approaching or exceeding human-level performance on a wider range of cognitive tasks.
The release strategy appears to involve staggered rollouts: an enhanced GPT-4o variant could arrive first for Pro and Enterprise users, followed by the full GPT-5-class model later in the year.
OpenAI has also been expanding its agent infrastructure, including the recent launch of “Operator” (a browser-based agent) and deeper integrations with tools like Canvas and Projects.
Industry observers expect the new models to feature larger context windows, better long-term memory, native tool-calling, and improved multimodal understanding. OpenAI has been training on massive clusters of Nvidia H100 and Blackwell GPUs, with compute costs reportedly in the billions.
The announcements come amid intense competition from Anthropic (Claude), Google (Gemini), and xAI (Grok), all racing to deliver more capable agents. OpenAI has not confirmed exact release dates or model names, and the company typically keeps final specifications under wraps until close to launch.
