Markopolo AI 发布 4B 参数大型事件模型 Athena（mvrko-sim-1），在 OPeRA 基准的购物行为预测任务中以 24.50% 严格精确匹配率超越 GPT-5.6、Claude Opus 4.8 等前沿模型。Athena 是 Qwen3-4B 的 LoRA 微调模型，可自托管，能预测购物者下一步浏览器动作及目标元素，为下游系统提供实时响应机会。
Newly launched 4B-param model Athena from @markopolo_ai just beat GPT-5.6 and Claude Opus 4.8 at predicting what a shopper does next!
So Athena watches the shopper’s current page plus their interaction history, then predicts the very next browser action.
e.g. when a shopper is about to leave the checkout journey, Athena can predict the exit action and the specific page element or navigation step involved even before it happens.
That could give a downstream system the opportunity to respond while the shopper is still active.
It scored 24.50% strict exact-match across the full OPeRA test set of 992 actions, the highest of every system entered.
Strict exact match means both the next action and the exact target element must be correct for the prediction to count; even a nearly correct prediction scores zero if the target element is wrong. This is a large, unforgiving output space and Athena still ranked first.
Model available on Huggingface.
Size is the surprise, since the winner is a LoRA fine-tune of Qwen3-4B, an open weight you can host yourself .
Until now the way to get a prediction like this was to prompt a frontier model and hope it worked out shopping behavior from general reasoning.
But this release is so significant, because the input can contain detailed browsing sequences, page state, typed text, and signals about what a user is trying to accomplish. Sending all of that to a general external model may be undesirable or too expensive at high event volume.
A compact, self-hostable specialist model offers a much more practical path.
Today, we’re releasing Athena (mvrko-sim-1), the flagship Large Event Model from @markopolo_ai that beats GPT-5.6, Claude Opus 4.8, and Claude Sonnet 5 in OPeRa public benchmark for Shopping behaviour prediction! Athena introduces a completely different approach to understanding digital behavior. Most systems record actions after they happen. Athena models the sequence behind them to predict what a shopper will do next and exactly where they will do it, before the action even takes place. Athena vs. frontier models: We evaluated Athena on the full OPeRA test set alongside GPT-5.6, GPT-4.1, Claude Sonnet 5, and Claude Opus 4.8. Athena achieved 24.5% strict exact match scoring the highest among the five models evaluated. Athena is not a general-purpose language model. It is a Large Event Model built to understand behavior as a connected sequence. It combines the page someone is viewing, the actions that brought them there, and the goal they are trying to complete and then predicts the exact next browser action and target element. That distinction matters. Today’s digital systems largely react after someone clicks, abandons, purchases, or leaves. Athena creates the foundation for systems that can understand what is likely to happen before the outcome. Together with @RubaiyatFarhan, and the team at @markopolo_ai, we built Athena from years of work on shopper behavior, digital journeys, and behavioral intelligence. Commerce was our first proving ground. Great thing is, we’re now releasing Athena publicly so builders can adapt the model to new datasets, industries, and problems. We’re seeing early usecases in Cyber Security, Gaming, Mobile App Ecosystem, Retail and more! Athena is now live on Hugging Face! We built the foundation for relevance and predictability! Now, let’s see what the world builds on top of it. Links in the first reply.在 X 查看被引用的帖子
来源：Rohan Paul · x.com