AI Paper Atlas

Research map / Large language models / Agents, code and math

ACPBench Hard: Unrestrained Reasoning about Action, Change, and Planning

Harsha Kokel, Michael Katz, Kavitha Srinivas, Shirin Sohrabi

ICLR 2026 · Large language models › Agents, code and math · approach: Reasoning benchmarks and platforms

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add PDDL representation to the NL prompt (ablation)DeepSeek V3 · ACPBench Hard avgaccuracy: 0.39 -> 0.47Improved
use PDDL only instead of NL (ablation)DeepSeek V3 · ACPBench Hard avgaccuracy: 0.39 -> 0.44Improved
use PDDL+NL instead of PDDL only (ablation)DeepSeek V3 · ACPBench Hard, just taskaccuracy: 0.74 -> 0.68Hurt
add PDDL+NL context (ablation)DeepSeek V3 · ACPBench Hard, val taskaccuracy: 0.56 -> 0.61No clear effect
use reasoning model o1-preview instead of GPT-4oo1-preview · ACPBench Hard, app taskaccuracy: 0.25 -> 0.44Improved
use reasoning model o1-preview instead of DeepSeek V3o1-preview · ACPBench Hard, just taskaccuracy: 0.65 -> 0.46Hurt
use reasoning model o1-preview instead of o1 minio1-preview · ACPBench Hard, val taskaccuracy: 0.78 -> 0.26Hurt
raise token cap to 4,000 for GPT OSS reasoning modelsGPT OSS 20B/120B · ACPBench HardaccuracyNo clear effect
scale from Llama 3.1 70B to 405BLlama 3.1 · ACPBench Hard, just taskaccuracy: 0.44 -> 0.59Improved
use generative format instead of boolean/MCQGPT-4o · ACPBench vs ACPBench HarderrorHurt
use less strict Jaccard scoring instead of exact set matcho1-preview · ACPBench Hard, app taskscore: 0.44 -> 0.57Improved
use reasoning models instead of LLMsOpenAI reasoning models · ACPBench Hard, areachaccuracy: 0.12No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy