AI Paper Atlas

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

AutoGPS: Automated Geometry Problem Solving via Multimodal Formalization and Deductive Reasoning

Bowen Ping, Minnan Luo, Zhuohang Dang, Chenxi Wang, Chengyou Jia

ICLR 2026 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation · 16 citations (OpenAlex)

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 pre-formalization (diagram and text parsers) before MLLM formalization (ablation)AutoGPS MPF · Geometry3K Completionaccuracy: +50.9% to +62.4%Improved
add multimodal alignment to pre-formalization (ablation)AutoGPS MPF · Geometry3Kaccuracy: +7.8% to +13.0%Improved
add multimodal alignment (ablation)InterGPS solver · Geometry3Kaccuracy: +4.2%Improved
add minimal solution generation (ablation)DSR · Geometry3Kaverage reasoning steps: 237.15 -> 16.69Improved
add geometry validation (ablation)DSR · Geometry3K (ground-truth formalization)Completion accuracy: 55.4 -> 94.5Improved
add geometry validation (ablation)AutoGPS (GPT-4o) · Geometry3KCompletion accuracy: 55.5 -> 75.4Improved
use both deductive and algebraic reasoning (ablation)DSR · Geometry3K (ground-truth formalization)Completion accuracy: 5.5 -> 94.5Improved
use feedback-based Refine@5 instead of Major@5 (ablation)AutoGPS · Geometry3K, PGPS9K, GeoQA Completionaccuracy: 66.3% -> 71.2% (Geo3K)Improved
use feedback-based Refine@5 instead of Pass@5 (ablation)AutoGPS · Geometry3K Completionaccuracy: 68.2% -> 71.2%Improved
AutoGPS framework vs best neural/symbolic SOTA on ChoiceAutoGPS (GPT-4o) · Geometry3K Choiceaccuracy: 82.3 -> 81.6No clear effect
AutoGPS framework vs best SOTA on ChoiceAutoGPS (GPT-4o) · PGPS9K Choiceaccuracy: 73.8 -> 81.5Improved
AutoGPS framework vs best SOTA on CompletionAutoGPS (GPT-4o) · Geometry3K and PGPS9K Completionaccuracy: 71.3 -> 75.4; 66.1 -> 75.3Improved

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