OpenAI Launches GPT-4.5 with Enhanced Reasoning Capabilities, Pricing at $75 per Million Input Tokens

Gate News message, April 16 — OpenAI unveiled its latest AI model, GPT-4.5, code-named Orion, featuring significantly increased computing power and training data compared to previous versions. The model was announced on Thursday.

ChatGPT Pro subscribers ($200/month) and research team members gain immediate access, while ChatGPT Plus and ChatGPT Team users will receive access around next week. For API developers, OpenAI is charging $75 per million input tokens and $150 per million output tokens, a substantial increase from GPT-4o’s $2.50 and $10 pricing respectively.

GPT-4.5 demonstrates improved performance on certain benchmarks, including SimpleQA for factual accuracy, and shows enhanced capabilities in understanding human intent with warmer, more natural responses and superior performance on creative tasks like writing and design. However, the model falls short of competitors’ reasoning models on advanced academic benchmarks such as AIME and GPQA, and lacks support for realistic two-way voice mode.

OpenAI noted that GPT-4.5 is a research preview, not a direct replacement for GPT-4o. The company plans to eventually merge its GPT series with its “o” reasoning series, beginning with GPT-5 expected later this year.