On Monday, OpenAI introduced the GPT-4.1 mannequin household, its latest sequence of AI language fashions that brings a 1 million token context window to OpenAI for the primary time and continues an extended custom of very complicated AI mannequin names. Three complicated new names, in actual fact: GPT‑4.1, GPT‑4.1 mini, and GPT‑4.1 nano.
In accordance with OpenAI, these fashions outperform GPT-4o in a number of key areas. However in an uncommon transfer, GPT-4.1 will solely be out there via the developer API, not within the client ChatGPT interface the place most individuals work together with OpenAI’s expertise.
The 1 million token context window—basically the quantity of textual content the AI can course of without delay—permits these fashions to ingest roughly 3,000 pages of textual content in a single dialog. This places OpenAI’s context home windows on par with Google’s Gemini fashions, which have provided related prolonged context capabilities for a while.
On the similar time, the corporate introduced it’s going to retire the GPT-4.5 Preview mannequin within the API—a brief providing launched in February that one critic known as a “lemon”—giving builders till July 2025 to modify to one thing else. Nevertheless, it seems GPT-4.5 will stick round in ChatGPT for now.
So many names
If this sounds complicated, effectively, that is as a result of it’s. OpenAI CEO Sam Altman acknowledged OpenAI’s behavior of horrible product names in February when discussing the roadmap towards the long-anticipated (and nonetheless theoretical) GPT-5.
“We understand how difficult our mannequin and product choices have gotten,” Altman wrote on X on the time, referencing a ChatGPT interface already crowded with selections like GPT-4o, varied specialised GPT-4o variations, GPT-4o mini, the simulated reasoning o1-pro, o3-mini, and o3-mini-high fashions, and GPT-4. The said purpose for GPT-5 will likely be consolidation, a branding transfer to unify o-series fashions and GPT-series fashions.
So, how does launching one other distinctly numbered mannequin, GPT-4.1, match into that grand unification plan? It is exhausting to say. Altman foreshadowed this sort of ambiguity in March 2024, telling Lex Friedman the corporate had main releases coming however was not sure about names: “earlier than we discuss a GPT-5-like mannequin known as that, or not known as that, or a little bit bit worse or a little bit bit higher than what you’d count on…”