FireWorldBench - Counterfactual Intervention Reasoning (Rendered Images): leaderboard
The counterfactual intervention reasoning dimension of FireWorldBench (Chen et al., 2026): predicting how a simulated fire would respond to an intervention such as a changed opening, exhaust, source or suppression. Observations are given as physics-rendered field images and 3D scene renderings; the score is completion accuracy in percent, averaged over matched multiple-choice and open-report questions and scored deterministically; higher is better.
Metric: Completion accuracy (%; P5, counterfactual intervention reasoning items of the controlled-simulation subset; image-observation track (physics-rendered field visualizations and 3D scene renderings); each cell is the mean of the matched choice-format score (Jaccard overlap of selected and gold option sets) and open-report score (exact agreement of required structured slots); deterministic evaluator, no LLM judge; failed or unparseable responses score 0). Source: arxiv.org. Saturation forecast: Around 2029. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 43.46 | #452 |
| 2 | InternVL3-78B | 41.22 | #733 |
| 3 | Qwen 3.5 Plus (2026-02-15) | 37.83 | #178 |
| 4 | Qwen 3 VL 8B Instruct | 36.42 | #982 |
| 5 | Qwen 3 VL 32B Instruct | 35.82 | #582 |
| 6 | GPT-5.6 Luna | 34.61 | #117 |
| 7 | Claude Haiku 4.5 | 32.47 | #568 |
| 8 | InternVL3-38B | 31.94 | #905 |
| 9 | InternVL3-8B | 26.27 | #1695 |
| 10 | Qwen 3 VL 235B A22B Instruct | 20.62 | #521 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=fireworldbench-counterfactual-intervention-reasoning-rendered-images · How It Works · Data refreshed daily, snapshot 2026-11-12.