OntoAct Reference - Mars (ReAct): leaderboard
Plain ReAct agent (Yao et al., 2023) row only, the paper's untrained baseline agent, compared across its three model conditions: GPT-5.6 Luna with reasoning effort none, GPT-5.6 Luna with medium effort and GPT-5.6 Terra with effort none. Mars (Tang et al., 2024) survival and crafting under altered world mechanics, eight-world configuration of Zhou et al. (2025); Score pools achievement success rates, mean over seven modified worlds. Methods share tasks and interaction budgets. Retrieval-augmented ReAct, Reflexion, IfR, WALL-E and OntoAct rows are agent methods and are not published. Metric: Mars Score (%). Higher is better. Source: https://arxiv.org/html/2610.21983v1
Metric: Mars Score (%). Source: arxiv.org. Saturation forecast: Around 2029. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra (Non-reasoning) | 13.91 | #41 (GPT-5.6 Terra) |
| 2 | GPT-5.6 Luna (Medium) | 13.06 | #68 (GPT-5.6 Luna) |
| 3 | GPT-5.6 Luna (Non-reasoning) | 7.95 | #68 (GPT-5.6 Luna) |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ontoact-reference-mars-react · How It Works · Data refreshed daily, snapshot 2026-10-18.