OpenAI has not officially announced GPT-5. But the clues — job postings, compute procurement, executive comments and the competitive landscape — paint a fairly clear picture of what the next generation of large language models will look like and when they will arrive.
Sam Altman, OpenAI’s CEO, has been characteristically cagey about timelines while making the direction clear. In a January 2025 interview, he described the jump from GPT-4 to the next model as “bigger than the jump from GPT-3 to GPT-4” — a significant claim, given that GPT-4 represented a roughly 10x increase in parameters (from 175 billion to an estimated 1.8 trillion) and dramatic improvements across virtually every benchmark. Altman also confirmed that OpenAI is training a next-generation model and that it requires a new training cluster.
The scale of that compute investment is staggering. OpenAI’s GPT-4 was trained on approximately 25,000 NVIDIA A100 GPUs. The next model is expected to use NVIDIA’s H100 or B100 GPUs — significantly more powerful — in clusters of 50,000 or more. Microsoft, OpenAI’s primary cloud partner, has been building out Azure infrastructure at a pace that suggests preparation for this scale. The company’s capital expenditures on cloud infrastructure reached $19 billion in a single quarter in late 2024.
What capabilities might GPT-5 have that GPT-4 lacks? The most likely advances are in reasoning (longer chains of logic without hallucination), multimodality (seamlessly processing text, images, audio and video as inputs and outputs) and tool use (autonomously calling APIs, browsing the web and executing code). Google’s Gemini 2.0, Anthropic’s Claude 3.5 and xAI’s Grok-3 provide benchmarks for what the state of the art looks like as of early 2025 — all are roughly GPT-4 class with incremental improvements rather than generational leaps.
One open question is whether GPT-5 will be “open-weight” in any meaningful sense. Meta’s Llama models have been released with publicly available weights (though restricted by a bespoke acceptable use policy), and the open-source AI community has rallied around them. OpenAI’s business model depends on selling API access to proprietary models, creating a fundamental tension with openness. How the company navigates this — particularly under pressure from competitors like Mistral and the broader open-source movement — will be one of the defining stories of AI in 2025 and 2026.