Dean Ball says "I do not expect DeepSeek to continue open sourcing their frontier models for all that much longer. I give it 12 months, max."
Is he correct?
This resolves YES if, after 3/23/25 and before 3/23/27 DeepSeek releases a model, and then within 3 months of that event releases the weights of that model, where the model in question is clearly their best publicly available model in terms of output quality, and can plausibly be called a 'frontier' model or a model that substantially advances the open weights SotA for output quality (regardless of cost considerations either way).
This resolves NO if, during that period, they release such a model and then does NOT release its weights within 3 months afterwards (I will not wait the 3 months if they make their intent clear here). This model does NOT need to plausibly be called a 'frontier' model or advance the open weights SoTA to count here.
This resolves to fair market price by my judgment, if by the close date (plus any extra time for the 3 month waiting period) neither the YES nor the NO condition is met.
If there is a dispute about how this market and I am genuinely unsure which way to resolve it, I will settle the outcome via a Twitter poll and abide by the results.
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ804 | |
2 | Ṁ275 | |
3 | Ṁ271 | |
4 | Ṁ150 | |
5 | Ṁ137 |
People are also trading
@TotalVerb DS themselves are calling it a 'minor trial upgrade' but it's too early to know if it is anything.
@ZviMowshowitz OS LLMs used to be ~10 percentage points behind the frontier closed source models in GPQA and HLA benchmarks, the gap has now closed to ~2 percentage points with the new R1. I don't think the wording of their announcement should influence the resolution here

Wow, that's quite the 'off by one' error on my part, so shifting by a year since what's done is done: https://manifold.markets/ZviMowshowitz/deepseek-opensource-frontier-model-dSs59QSg5y?play=true