The model must be able to successful generate an image of a horse riding an astronaut on the moon with only the exact string "image of a horse riding an astronaut on the moon" at least 5/10 times. If the model doing the generation is multimodal "Generate an" will be prepended to the string.
https://x.com/fofrAI/status/1851661066566316168 According to this X poster, Sama promised big improvements to image models soon (possibly related to o1 like test-time compute scaling).
Update 2025-12-27 (PST) (AI summary of creator comment): The creator will conduct their test run now (before the market close date) using gpt image 1.5, rather than waiting until closer to 2026.
Update 2025-12-27 (PST) (AI summary of creator comment): The creator has clarified that for an image to count as a pass, the horse must be riding the astronaut (not the astronaut riding the horse). An image showing an astronaut riding a horse will be counted as a fail.
Update 2025-12-28 (PST) (AI summary of creator comment): Modified prompts are not allowed. The model must generate the correct image using only the exact prompt specified in the description, without any additional formatting, punctuation, or modifications (such as adding dashes or line breaks).
1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ271 | |
| 2 | Ṁ173 | |
| 3 | Ṁ168 | |
| 4 | Ṁ110 | |
| 5 | Ṁ107 |
People are also trading
gave 5 stars, but I personally disagree that openAI can't generate that image.
chatGPT is in fact capable, more than capable, of generating the exact image, including a very good depiction of a horse riding on a person’s back. The issue is that the prompt itself is designed to mislead ChatGPT, while some other models do not attempt to correct the potential ambiguity in the initial prompt. I believe ChatGPT can generate the intended image if we explicitly tell OpenAI that we need a horse riding on a person, not the opposite.
i gave description of this market, from the exact prompt chatGPT: https://chatgpt.com/share/6950681d-c08c-8010-92f3-1c874833867f
or exact prompt: https://chatgpt.com/share/695069c9-7bd4-8010-88ea-207a8732170d
Since I doubt OpenAI will release a new image model between now and then, I will do my run now (on gpt image 1.5) and report results
UPDATE: GPT-image-1.5 (on the api) went 0/10. I will try GPT-image-1 next to be thorough (and because I still have API credits left), but I doubt the results will be much different
UPDATE 2: GPT-image-1 also went 0/10
@JaundicedBaboon can you try to put a pause break in between the horse and the astronaut because I don’t think your script will ever give you the correct result because of the meaning of the sentence itself.
If you try putting it like this:
image of a horse - riding an astronaut on the moon
The script stays the same, but it makes it clear what would be the subject of the picture
@MindBenderMads I think that's a good idea, but my market set clear criteria for what the prompt would be. Maybe I'll make a new market for a different prompt
@JaundicedBaboon actually ChatGPT 5.2 do it the correct : https://chatgpt.com/share/6950681d-c08c-8010-92f3-1c874833867f
@JaundicedBaboon May I ask why you work with older OpenAI models and don't try the latest 5.2? It generates images well, exactly with a horse on the back of an astronaut. Exact prompt: https://chatgpt.com/share/695069c9-7bd4-8010-88ea-207a8732170d
@1bets this does not meet the criteria because the model is given additional prompting that is not allowed.
I think it’s a notable success, but it nonetheless cannot resolve yes
@pureprofit the question mentions the exact prompt that should be used, and I personally couldn’t get it to work with a one liner
5.2 achieved (from the 2nd attemp https://chatgpt.com/share/694352ce-3bac-8010-beaf-dde49ea4f7e9)
