GPT-4.5 preview

[simonw]

https://openai.com/index/introducing-gpt-4-5/

simonw/scrape-openai-models@018e600 shows model IDs:

• gpt-4.5-preview-2025-02-27
• gpt-4.5-preview

[simonw]

This is surprising:

llm -m gpt-4.5-preview-2025-02-27 'tell me your training cutoff'

My knowledge cutoff date is October 2023. This means I don't have information about events or developments occurring after that time.

I wonder if that's correct?

[simonw]

llm -m gpt-4.5-preview 'Generate an SVG of a pelican riding a bicycle'

<svg xmlns="http://www.w3.org/2000/svg" width="300" height="250" viewBox="0 0 300 250">
  <!-- Bicycle Wheels -->
  <circle cx="90" cy="200" r="40" stroke="#000" stroke-width="4" fill="none"/>
  <circle cx="210" cy="200" r="40" stroke="#000" stroke-width="4" fill="none"/>
  
  <!-- Bicycle Frame -->
  <line x1="90" y1="200" x2="150" y2="150" stroke="#000" stroke-width="4"/>
  <line x1="150" y1="150" x2="210" y2="200" stroke="#000" stroke-width="4"/>
  <line x1="150" y1="150" x2="150" y2="120" stroke="#000" stroke-width="4"/>
  <line x1="90" y1="200" x2="130" y2="130" stroke="#000" stroke-width="4"/>
  <line x1="130" y1="130" x2="180" y2="130" stroke="#000" stroke-width="4"/>
  <line x1="210" y1="200" x2="180" y2="130" stroke="#000" stroke-width="4"/>
  
  <!-- Bicycle Handlebar -->
  <line x1="180" y1="130" x2="190" y2="115" stroke="#000" stroke-width="4"/>
  <line x1="180" y1="130" x2="170" y2="115" stroke="#000" stroke-width="4"/>
  
  <!-- Pelican Body -->
  <ellipse cx="150" cy="90" rx="30" ry="40" fill="#fff8e7" stroke="#000" stroke-width="3"/>
  
  <!-- Pelican Neck -->
  <path d="M150 60 Q170 10 200 40" fill="none" stroke="#000" stroke-width="3"/>
  
  <!-- Pelican Head -->
  <ellipse cx="200" cy="40" rx="15" ry="20" fill="#fff8e7" stroke="#000" stroke-width="3"/>
  
  <!-- Pelican Beak -->
  <polygon points="215,45 250,55 210,65 200,50" fill="#f5c55f" stroke="#000" stroke-width="3"/>
  
  <!-- Pelican Eye -->
  <circle cx="205" cy="35" r="3" fill="#000"/>
  
  <!-- Pelican Wing -->
  <path d="M130 80 Q100 70 120 100 Q130 90 130 80" fill="#e8dcc0" stroke="#000" stroke-width="2"/>
</svg>

[simonw]

Need to make sure I'm supporting the right features. https://platform.openai.com/docs/models#gpt-4-5 says:

This is a research preview of GPT-4.5, our largest and most capable GPT model yet. Its deep world knowledge and better understanding of user intent makes it good at creative tasks and agentic planning. It accepts both text and image inputs and produces text outputs (including Structured Outputs). GPT-4.5 also supports key developer features like function calling, Batch API, and streaming.

[simonw]

It's pretty slow:

[simonw]

Until there's a release out you can test this like so:

uvx --with 'https://github.com/simonw/llm/archive/801b08bf40788c09aed6175252876310312fe667.zip' \
  llm -m gpt-4.5-preview 'impress me, somehow'

[simonw]

It's very expensive! https://openai.com/api/pricing/ - currently $75.00 / 1M tokens input and $150/1M output! For comparison o1 is $15/$60 and GPT-4o is just $2.50/$10.

[simonw]

Blogged a bunch of notes here: https://simonwillison.net/2025/Feb/27/introducing-gpt-45/

[simonw]

Given OpenAI's lack of confidence that they'll be releasing a full model:

GPT‑4.5 is a very large and compute-intensive model, making it more expensive⁠ than and not a replacement for GPT‑4o. Because of this, we're evaluating whether to continue serving it in the API long-term as we balance supporting current capabilities with building future models.

I'm going to assign the gpt-4.5 alias to this preview model. I may live to regret that slightly.