Upcoming o4-mini and o3 models hint at next phase of OpenAI’s AI lineup

References to three potentially upcoming OpenAI models – "o4-mini," "o4-mini-high," and "o3" – were recently discovered in the code of the ChatGPT web application by Tibor Blaho and confirmed by TestingCatalog. This discovery suggests OpenAI is preparing to expand its lineup of AI models available through ChatGPT.

These new models would join existing options like GPT-4o, o1, and the current o3-mini variants. The "o3" reference likely points to the full version of the o3 reasoning model, succeeding o1; currently, only the smaller o3-mini and o3-mini-high versions are available. The o3 series focuses on strong performance in STEM fields (science, math, coding) and logical reasoning, offering cost-effective alternatives to larger models. The "o4-mini" and "o4-mini-high" names suggest a next generation of smaller reasoning models, possibly featuring adjustable "reasoning effort" levels similar to o3-mini, allowing users to balance performance and speed. These models would likely appear within the ChatGPT interface, selectable by users depending on their subscription tier and task requirements. Beneficiaries would include ChatGPT users seeking different balances of capability, speed, and cost, particularly developers and those working on STEM-related problems.

While no official release date is confirmed, OpenAI CEO Sam Altman previously indicated plans to release new o3 and o4 models "in a couple of weeks" (from his post date) ahead of the larger GPT-5 model, citing development progress. The discovery of these references in the deployed web app indicates preparations are actively underway. This potential release aligns with ongoing competition, notably Google's recent advancements with Gemini 2.5 Pro and other announcements during the Google Cloud Next event.

New Models

The identified models represent an expansion of OpenAI's "o-series" reasoning models.

1. o3 (full): Expected to be the complete successor to the o1 reasoning model, offering advancements over the currently available o3-mini versions. The o3 family is optimized for STEM tasks, cost-efficiency, and lower latency compared to larger models. It reportedly achieved high scores on reasoning benchmarks like ARC-AGI.
2. o4-mini / o4-mini-high: These likely represent the next step in the reasoning series, presumably offering improved reasoning capabilities over the o3 generation. The "mini" and "mini-high" naming, mirroring the o3-mini structure, suggests different performance tiers possibly linked to adjustable "reasoning effort," a feature allowing users to trade speed for analytical depth. Limitations compared to full models might exist (e.g., o3-mini lacks the vision capabilities of o1).

About OpenAI

OpenAI maintains a strategy of releasing diverse ChatGPT models catering to different needs, including general-purpose models (like GPT-4o) and specialized reasoning models (the o-series). They often introduce "mini" versions first, providing faster, lower-cost options, particularly strong in areas like coding and logic. These upcoming models fit into OpenAI's tiered product structure (free, plus, pro, enterprise) and reflect a continuous push to advance reasoning capabilities, potentially supporting more complex agentic tasks and enterprise applications in the future. The timing also suggests a competitive response to maintain leadership in the AI space.