Sam Altman has signaled that he will open-source o3-mini, or perhaps an even smaller model which can run on phones. He has also indicated that OpenAI might adopt a broader policy of releasing mini versions of successive generations of models.
Conditional on OpenAI releasing the weights of a model based on o3-mini, this market resolves YES if OpenAI releases the weights of a model broadly more powerful model than o3-mini before July 2026. Note that if OpenAI releases a distilled version of o3-mini, a qualifying next model must be better than o3-mini itself, not just the distillation.
My criterion for “more powerful” is roughly “a pareto improvement across most benchmarks.” Successive generations of models tend to be better in this way, so I expect this to be clear to judge. However, in the case of ambiguity I will defer to a panel of 3 mods.
1,000People are also trading
@mods please resolve this market to "no". No new open-weights models have been released by OpenAI.
o3-mini is matched by gpt-oss 120b on all benchmarks reported by OpenAI in their blog post so this market is valid: https://openai.com/index/introducing-gpt-oss/
Both gpt-oss models were also released at the same time and on the same day so this market doesn't resolve to "yes" on a technicality.
@AdamK how does this resolve? I vote for "no", since OpenAI hasn't released another series of open-weight models since gpt-oss
@Bayesian It’s so easy. You give them a taste and they come scrambling back. I will say that if the model is like <8B parameters and very good I will be shamelessly using it for everything.
