OpenAI has launched two very expensive language models recently (GPT 4.5, o1-pro). If they launch any language model in 2025 that costs over $600 per million tokens (input or output) in their API this market resolves YES.
This will be based off numbers as shown on the OpenAI API pricing page, I may ignore a new model if it has unusual pricing that makes it difficult to compare.
Update 2025-03-21 (PST) (AI summary of creator comment): Clarification on Ignored Models and Resolution Outcomes
Ignored Models: If a model is ignored due to unusual pricing, it cannot trigger a YES outcome.
Outcome: The market will not be marked as N/A; if no qualifying model is launched, the resolution is NO.
Update 2025-06-04 (PST) (AI summary of creator comment): - The creator will evaluate a model's cost based strictly on the listed price on the OpenAI API pricing page, disregarding model internals or mechanics like hidden tokens.
A model with variable pricing (e.g., different prices for different context window sizes) will be considered "difficult to compare" and will not trigger a YES resolution if its pricing straddles the $600/MTok threshold (i.e., some tiers are below $600/MTok while others are above).
Example: A model priced at $400/MTok for small contexts and $800/MTok for large contexts would not count towards a YES resolution.
If all pricing tiers for such a variable-priced model are strictly above $600/MTok, it will not be considered "difficult to compare" and will count towards a YES resolution if it otherwise qualifies.
Example: A model priced at $800/MTok for small contexts and $1000/MTok for large contexts would count.
1,000People are also trading
it's weird bc with o1-pro almost all generated token is not shown, and you only pay for tokens shown, so it appears like it's $600 per million tokens generated but it isn't actually. that will likely continue to be the case, so i'm wondering if you could confirm that similar things would still count, and not count as being "difficult to compare" or wtv, bc that might end up being relevant. as far as I'm concerned o1-pro was already difficult to compare with everything before it
@Bayesian I’m just going to look at pricing and not care about model internals / hidden outputs. If it’s a text generation model and the price page says it costs over $600 per million tokens, it counts.
What I mean by difficult to compare is stuff like Gemini, where the price per token varies depending on context window size. If they were to charge $400/MTok for small contexts and $800 for large, I would say that’s hard to compare because it depends on the circumstances if it’s more expensive per token, so it doesn’t count. If it’s $800 for small and $1000 for large, that’s not hard to compare, it’s strictly more expensive,
I may ignore a new model if it has unusual pricing that makes it difficult to compare.
Just to confirm: To ignore such a model would cause this to resolve No? Or N/A?
@Primer If I ignore a model it just means it can’t cause the market to resolve YES. I don’t plan on this market N/Aing