# Luzmo updates World Cup AI predictor with scenario prompts > Source: > Published: 2026-06-13 14:18:34.998318+00:00 # Luzmo updates World Cup AI predictor with scenario prompts The team behind the AI Octopus Euro 2024 predictor has updated its simulator for the 2026 FIFA World Cup to accept natural-language scenarios, per The Register. Users can submit sensible or silly prompts, including hypotheticals such as a red card, a heat wave, or even "What if the tournament were played with rugby rules?", a quote attributed to Luzmo CTO Haroen Vermylen in The Register. The system uses squad-quality data, heat and altitude factors, injury inputs and a Monte Carlo engine that derives scorelines from **5,000** match runs, according to The Register. The implementation moved from TypeScript to **Rust** to cut latency; Vermylen told The Register predictions previously took about five minutes and now should return within "two to three seconds of actual simulation time." OpenAI models parse prompts and an agent orchestrates scenario creation, calculation calls and summaries, and filtering blocks profanities and prompts "harmful to certain groups," Vermylen told The Register. The Register reports an early baseline puts **Spain** at an **18 percent** chance to lift the trophy and **26.8 percent** chance of reaching the final. ### What happened Per The Register, the team behind the AI Octopus Euro 2024 predictor released an updated simulator for the 2026 FIFA World Cup that accepts freeform natural-language scenarios. "Sensible questions work ... but so do the daft ones," Luzmo CTO Haroen Vermylen told The Register, using the example "What if the tournament were played with rugby rules?" The model combines squad-quality data, heat and altitude factors, injury inputs and a Monte Carlo simulation; The Register reports scorelines are derived from **5,000** match runs. The Register also reports the engine was rewritten from TypeScript into **Rust** to improve responsiveness, and Vermylen told The Register the team aims to cut runtime from about five minutes to "two to three seconds of actual simulation time." OpenAI models parse user requests and an agent handles scenario transformation, calls to the calculation engine and natural-language summaries. Vermylen told The Register that filtering is in place to ignore profanities and "to avoid scenarios that would just be harmful to certain groups." The Register gives a baseline indicating **Spain** has an **18 percent** chance to win the trophy and a **26.8 percent** chance of reaching the final. ### Technical details Per The Register, the predictor uses a Monte Carlo approach over thousands of simulated matches to generate win/lose/draw probabilities and scorelines. The move to **Rust** was reported as a runtime optimization to support the real-time component; OpenAI models are used for prompt parsing and an agent coordinates scenario generation and result presentation. The system ingests player-level and contextual features such as heat and altitude, according to The Register. ### Industry context Editorial analysis: Interactive, natural-language scenario interfaces are an increasingly common front end for simulation systems because they lower the barrier for non-experts to pose counterfactuals. For practitioners, this pattern highlights tradeoffs between model fidelity, input sanitization and latency engineering when simulations are exposed in near real time. Systems that combine external LLMs with an internal simulation engine typically require careful orchestration to avoid prompt misunderstanding and to enforce content safety controls. ### What to watch Monitor latency and cost metrics as the team adds live data; observe how scenario sanitization handles ambiguous or adversarial prompts; and watch for published validation or calibration tests showing how well the Monte Carlo outcomes match historical results. Also watch whether the team publishes technical notes on feature engineering for environmental factors like heat and altitude. ## Scoring Rationale This is a notable applied demo combining LLM-driven prompts with a Monte Carlo simulator and real-time optimizations. It matters to practitioners building interactive simulation tooling but is not a frontier research release. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)