AI Model Pricing: The Leap from GPT-4o to 4.5
Description
A screenshot of a pricing comparison table for two AI models, presented in a dark mode UI. The table has columns for 'Model', 'Input', 'Cached input', and 'Output'. The first row lists 'gpt-4.5-preview' with a date of 2025-02-27, showing costs of $75.00 for input, $37.50 for cached input, and a staggering $150.00 for output. The second row lists 'gpt-4o' with a date of 2024-08-06, with significantly lower costs: $2.50 for input, $1.25 for cached input, and $10.00 for output. This image is a meme that satirizes the rapid and often extreme escalation in the cost of next-generation AI models. It humorously projects a future where accessing a preview of a slightly more advanced model is exponentially more expensive, a relatable concern for developers and companies who rely on these APIs and must manage their costs carefully
Comments
18Comment deleted
The gpt-4.5-preview's price is so high because it includes the cost of the therapist you'll need after you see the first invoice
A single line in the model-name field: from gpt-4o to gpt-4.5-preview - never seen YAML turn into a board-level spending request so fast
Remember when we complained about AWS egress charges? Now we're paying $150 per million output tokens and calling it 'investing in innovation' while the CFO quietly updates their LinkedIn to 'open to opportunities'
When your CFO sees the GPT-4.5-preview pricing and suddenly becomes very interested in 'technical debt reduction' by sticking with GPT-4o. Nothing says 'bleeding edge' quite like a 30x price multiplier - at $150 per million output tokens, that's not a language model, that's a subscription to having an existential crisis every time someone hits your API endpoint. Suddenly that 'we should cache everything' architecture decision doesn't seem so premature after all
Those prices made prompt caching an architecture principle - we call it CFO‑driven design: POC on gpt‑4.5, production quietly on gpt‑4o
We implemented Mixture of Expenses: a prompt router uses 4o for 99.9% of tokens and only trips to 4.5 behind a rate-limited circuit breaker with a CFO webhook
Cached prompts in GPT-4: Because re-tokenizing your architecture doc should cost full price every time
What am i looking at? Comment deleted
https://openai.com/api/pricing/ Comment deleted
Pricing for gpt 4.5 per 1M tokens Comment deleted
a joke Comment deleted
Garbage Comment deleted
why would they even release this, like its not even close to some game - changer but costs 10 godzillion dollars Comment deleted
hopefully Comment deleted
Thank God I never became an aibro Comment deleted
They really want us to use the reasoning models Comment deleted
It reminds me about one of those mobile ads. Comment deleted
Journalist that has lower price per word: 💀 Comment deleted