GPT-5 Delayed but Will Be Free, OpenAI Plans Open‑Source Model; Meta’s Llama 4 Continues to Be Postponed

OpenAI’s spokesperson, nicknamed "Otterman," revealed that the development of GPT‑5 is more challenging than expected, leading to a delay of several months, but the model will eventually be offered for free to users.

In the meantime, OpenAI promises the near‑term launch of o3 and o4‑mini within weeks, and hints at a mysterious upcoming open‑source inference model that may be the next iteration of the 4o series.

OpenAI also announced that GPT‑4.5 (Orion) will be the last non‑chain‑of‑thought model, and that GPT‑5 will unify all tools and functions, eliminating separate releases of o3. Free users will have unlimited access, while paid users can use a higher‑intelligence version.

According to the report, GPT‑5 will integrate voice, Canvas, search, Deep Research and other capabilities into a single system that can decide when to think deeply and when to respond quickly, aiming to simplify the model and product ecosystem.

OpenAI plans to open‑source a powerful inference model in the coming months, the first such release since GPT‑2, potentially making GPT‑5 itself open‑source.

Meta’s Llama 4, once a leading open‑source model, continues to be delayed. Reports indicate the model has not met benchmark expectations in reasoning and mathematics, and may lag behind OpenAI’s conversational abilities.

Meta is reportedly preparing to release Llama 4 later this month, but further postponements are possible as the company invests heavily—up to $650 billion—in AI infrastructure and considers incorporating DeepSeek’s mixed‑expert (MOE) technology.

Overall, the AI community anticipates that free access to GPT‑5 and the upcoming open‑source model will reshape the competitive landscape, while smaller firms may struggle against the resource‑intensive strategies of major players.