A Look Under the Hood: How ManticAI Orchestrates a Team of LLMs

The impressive eighth-place finish of ManticAI in a forecasting competition was not the work of a single “genius” AI, but of a masterful conductor orchestrating a whole orchestra of them. A look under the hood reveals a sophisticated system that gets its power from expertly managing a team of large language models (LLMs) from different developers.
The startup’s core innovation is this orchestration layer. ManticAI “breaks down a forecasting problem into different jobs” and then assigns those jobs to a “roster of machine-learning models including OpenAI, Google and DeepSeek, depending on their strengths.” This is the secret to its success.
This multi-agent, multi-model architecture is highly efficient. One model, perhaps with strong analytical capabilities, might be tasked with assessing current data. Another, with more “creative” reasoning, could be used to game out future scenarios. A third, optimized for speed, might be responsible for continuous monitoring of news feeds.
By using the best tool for each part of the job, the system as a whole becomes incredibly powerful and versatile. It can tackle a wide range of questions, from Samoan elections to US wildfires, by dynamically assembling the right team of AI agents for each specific challenge.
This “orchestration” approach represents the next wave of AI development. It’s a shift from just building bigger models to building smarter systems that can combine and leverage the capabilities of existing models. ManticAI’s success is a powerful testament to the effectiveness of this new paradigm.