OpenAI Readies GPT-4.1, o3, o4-mini Launch Expected Next Week

OpenAI appears set to introduce a suite of new artificial intelligence models, potentially making them available as early as next week, according to individuals familiar with the company’s roadmap cited by The Verge. This expected rollout includes an updated multimodal model tentatively named GPT-4.1, described as a successor to GPT-4o which processes text, audio, and vision, along with scaled-down ‘mini’ and ‘nano’ counterparts.

Concurrently, the AI lab is reportedly finalizing its specialized reasoning models, o3 and o4-mini, for near-term release. Adding weight to these reports, AI engineer Tibor Blaho posted on X today, that he discovered code references to o3, o4-mini, and o4-mini-high within a just-deployed ChatGPT web update. The Verge noted that OpenAI did not respond to their request for comment before publication.

Strategy Shift Delays GPT-5

These imminent releases signal a course correction in OpenAI’s product strategy. On April 4th, CEO Sam Altman confirmed a “Change of plans”, stating the intention to release o3 and o4-mini “probably in a couple of weeks,” pushing the launch of the widely anticipated GPT-5 back by “a few months.” He attributed this partly to a decision to “decouple reasoning models and chat/completion models.”

This marks a departure from OpenAI’s February plan to absorb the o3 model’s functions directly into GPT-5, a move previously aimed at simplifying its product lineup. Altman added regarding the upcoming o3 release, “we are excited about the performance we’re seeing from o3 internally.” He also noted on X, “there are a bunch of reasons for this, but the most exciting one is that we are going to be able to make GPT-5 much better than we originally though[t]”

The renewed focus on separate reasoning models like o3 addresses a capability gap observed even in OpenAI’s recent updates. When GPT-4.5 was launched in late February as an efficiency and multilingual improvement over GPT-4o, the company’s own GPT-4.5 system card acknowledged it underperformed specialized models like o3-mini on certain complex tasks demanding rigorous logic. Altman described GPT-4.5 at the time as feeling like talking to a thoughtful person, but conceded it wasn’t designed as a “reasoning powerhouse,” setting the stage for the separate release of models like o3.

The upcoming full o3 model, initially previewed in December 2024, is expected to excel in these areas, leveraging techniques like private chain-of-thought (internal logical steps before output) to tackle complex problems. Benchmarks from its preview showed high scores in mathematics (96.7% on AIME 2024) and science reasoning tests (87.7% on GPQA Diamond). However, this advanced reasoning comes with potentially steep compute requirements; benchmark reports indicated usage up to 172 times higher in certain configurations.

The expected o4-mini and the discovered o4-mini-high likely represent different tiers within this specialized reasoning family, possibly offering similar benefits with lower resource demands, though specifics are not yet public.

GPT-4 Line Update Amid Market Pressures

Alongside the specialized ‘o’ series, the report points to GPT-4.1 as the next iteration in OpenAI’s main multimodal model line, building upon GPT-4o. The existence of planned ‘mini’ and ‘nano’ versions suggests a tiered approach, similar to previous generations, offering different balances of capability and resource demand for the core GPT model family. OpenAI’s launch plans enter a market where access to sophisticated AI reasoning is varied. The company already markets the high-cost, enterprise-focused o1-Pro model via API, which OpenAI documentation states “uses more compute to think harder and provide consistently better answers.”

In contrast, partner Microsoft integrated the o3-mini-high model into its free Copilot tier in early March. The positioning of the full o3 and the o4-mini series within OpenAI’s own paid structure remains unclear.

While the launch seems close, potential delays persist. Sources cited by *The Verge* noted capacity constraints could shift the timeline. Altman himself addressed this earlier in April, posting that users “should expect new releases from OpenAI to be delayed, stuff to break, and for service to sometimes be slow as we deal with capacity challenges.” These constraints arise despite OpenAI’s massive funding rounds and major investments in compute resources, including an $11.9 billion deal with CoreWeave and participation in the government-backed Stargate Project for domestic AI infrastructure.