Archive — US — techcrunch
OpenAI unveils GPT-4.5 ‘Orion,’ its largest AI model yet
OpenAI launched its largest AI model, GPT-4.5 'Orion', which utilizes significant computational power and data but is not deemed a frontier model. Available first to ChatGPT Pro subscribers, it outperforms GPT-4 in certain tasks but faces competition from newer reasoning models. OpenAI acknowledges the financial viability of GPT-4.5 may be uncertain despite its enhanced features and emotional intelligence.
Key points
- OpenAI unveiled GPT-4.5 on Thursday, the largest model to date, using unprecedented computing power and data.
- GPT-4.5 is available to ChatGPT Pro subscribers at $200/month as a research preview.
- While it shows improved performance over GPT-4 in factual accuracy and emotional intelligence, it lags behind newer models from competitors including DeepSeek and Anthropic.
- Challenges faced with scaling models indicate that traditional scaling laws in AI development may not hold, leading to a shift towards reasoning models.
- GPT-4.5's high operational costs raise questions about its long-term availability in the API.
Why it matters
- The launch of GPT-4.5 comes amid increasing competition in the AI landscape, particularly from companies focused on reasoning-centric models.
- Historically, OpenAI has consistently leveraged larger datasets and computing for model development, a trend that is now being scrutinized for sustainability.
- The growing concern over scaling laws reflects broader industry trends toward refining AI models for greater efficiency and consistency.
GPT-4.5 represents a significant but cautious step in AI development, emphasizing the need to adapt to shifting technological demands and competitive pressures as the industry grapples with the limitations of traditional training approaches.