OpenAI Releases GPT-4.5

OpenAI released GPT-4.5 on Thursday, following the launches of Anthropic’s Claude 3.7 Sonnet and xAI’s Grok-3.

Pricing

GPT-4.5 comes with a notably high API price tag of $75 per million input tokens and $150 per million output tokens, making it ten times pricier than Claude 3.7 Sonnet. To put this into perspective, it is 2900% more expensive for input tokens and 1300% more for output tokens compared to its predecessor, GPT-4o.

CEO Remarks

Sam Altman, OpenAI’s CEO, highlighted the massive resources required for the model, stating that it is not a reasoning model but represents a new type of intelligence. He described moments of being astonished by the quality of advice received from the AI.

User Experience

According to Altman, GPT-4.5 focuses on the ‘vibes,’ enhancing emotional intelligence and collaborative interaction. OpenAI created a “Vibes test set” to assess the model’s conversational quality, which showed improved nuances, even though some demos lacked innovation.

Demonstrations

The initial demo prompted the AI with an exaggerated expression from a user upset about a friend canceling plans. The response showed tone improvement but not significant breakthroughs. For example, when asked why seawater is salty, GPT-4.5 opted for simpler language.

Performance Benchmarks

GPT-4.5 scored 71.4% on the GPQA evaluation, higher than its predecessor’s score (53.6%) but still below OpenAI’s o3-mini model (79.7%). The model performed well on some coding benchmarks but didn’t consistently outperform reasoning-focused models.

Development Innovations

To develop GPT-4.5, OpenAI implemented significant technical innovations, including new inference systems, low-precision training methods, and training across multiple data centers to handle the large model’s requirements.

Availability

GPT-4.5 is currently available for Pro users at $200 a month, with Plus users gaining access the following week.

Edited by Sebastian Sinclair