Papers with FIGR
Figure It Out: Improve the Frontier of Reasoning with Executable Visual States (2026.acl-long)
Copied to clipboard
| Challenge: | Recent reasoning models fail to capture structural constraints in complex settings. |
| Approach: | They propose a visual-based reasoning system that integrates executable visual construction into multi-turn reasoning via end-to-end reinforcement learning. |
| Outcome: | The proposed model outperforms strong text-only chain-of-thought models on seven mathematical benchmarks and improves by 13.12% on AIME 2025 and 11.00% on BeyondAIME. |