OpenAI has launched GPT-4.5, its latest large language model, boasting significantly reduced “hallucinations” compared to its predecessor, GPT-4o. This release comes amidst a flurry of new AI model announcements from major tech companies, intensifying competition within the rapidly evolving AI landscape.

The San Francisco-based company revealed that GPT-4.5 achieved a 37% hallucination rate in early tests, a substantial improvement over the nearly 60% rate observed in GPT-4o. This focus on accuracy underscores OpenAI’s continued investment in large, powerful AI models, even as smaller, more accessible alternatives emerge.

“With every new order of magnitude of compute comes novel capabilities,” OpenAI stated in a blog post, emphasizing GPT-4.5’s “broader knowledge and a deeper understanding of the world.” The company also highlighted the model’s increased reliability across various topics.

The launch of GPT-4.5 follows recent releases from competitors, including Anthropic’s Claude 3.7 Sonnet and Elon Musk’s xAI’s Grok 3, demonstrating the accelerated pace of innovation in the AI sector.

OpenAI, a leader in the global AI race, is reportedly in talks to raise up to $40 billion at a $300 billion valuation. Similarly, Anthropic is seeking $3.5 billion at a valuation exceeding $60 billion, reflecting the substantial capital flowing into the AI industry.

GPT-4.5 will initially be available as a preview to developers through OpenAI’s API, though the company has indicated that long-term access is contingent on usage and cost considerations. “GPT‐4.5 is a very large and compute-intensive model, making it more expensive than and not a replacement for GPT‐4o,” OpenAI explained.

The high cost of developing and running such large models is a significant factor. OpenAI CEO Sam Altman previously disclosed that GPT-4 cost over $100 million to train, and these expenses are expected to rise with future model advancements.

In a post on X, Altman acknowledged the company’s current GPU shortage, a critical resource for AI model training and operation.

“This isn’t how we want to operate, but it’s hard to perfectly predict growth surges that lead to GPU shortages,” he wrote, adding that they expect to receive more GPUs soon. He also noted that while GPT-4.5 might not excel in traditional reasoning benchmarks, it possesses “a different kind of intelligence.” Sources and related content