Olas Unveils Smaller AI Model That Matches GPT-4.1 in Prediction Accuracy
Olas, a company developing autonomous AI agents for prediction markets, has released an AI model called Olas-Predict-R1-14B that matches the forecasting accuracy of OpenAI's GPT-4.1 while using significantly fewer resources. The model achieved 75.8% accuracy in predicting real-world events across over 2,600 unseen markets—nearly equal to GPT-4.1’s performance and outperforming its base DeepSeek model. David Minarsch of Valory and a founding member of Olas said the results highlight how smaller, specialized AI models can challenge larger general-purpose systems by offering comparable accuracy at lower costs. This could shift economic dynamics in the AI industry as companies like NVIDIA benefit from demand for high-end computing infrastructure driven by large models. The emergence of efficient specialist models may reduce reliance on expensive frontier systems and impact capital spending trends among cloud providers, which are projected to spend over $886 billion this year on AI-related infrastructure. Olas-Predict is publicly available and can run on a single GPU.
Topics: AI NVIDIA Olas OpenAI prediction models