{"paper":{"title":"GeoBuildBench: A Benchmark for Interactive and Executable Geometry Construction from Natural Language","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Current multimodal models often produce geometry diagrams with hallucinations, missing objects, and violated constraints from natural language problems.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huishuai Zhang, Jinwoong Kim, Rui Yang","submitted_at":"2026-05-13T08:30:12Z","abstract_excerpt":"We introduce GeoBuildBench, a benchmark designed to evaluate whether large language models and multimodal agents can ground informal natural-language plane geometry problems into executable geometric constructions. Unlike existing geometry benchmarks that focus on answer correctness or static diagram interpretation, GeoBuildBench treats geometry diagram as an interactive construction task: given a textual problem, an agent must generate a domain-specific language (DSL) program to produce a diagram satisfying explicitly specified geometric objects and verifiable constraints. The benchmark featu"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"despite reasonable success rates, models frequently exhibit structural hallucinations, missing objects, and failures to satisfy geometric constraints, with limited ability to exploit visual and constraint-based feedback for self-correction","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 489 problems are text-complete and constructible after automated filtering and human validation, and the chosen DSL accurately represents valid geometric constructions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GeoBuildBench is a new benchmark requiring LLMs to generate executable geometry constructions from text, revealing frequent hallucinations, missing objects, and constraint failures in state-of-the-art models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Current multimodal models often produce geometry diagrams with hallucinations, missing objects, and violated constraints from natural language problems.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7700514e8b32e1fa170da2b41b42d4bcbb6ce3e57bbf280c7dc20c094e4bdfac"},"source":{"id":"2605.13167","kind":"arxiv","version":1},"verdict":{"id":"2db43c02-abd6-48cb-8cbe-5ff76b0b110f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:17:00.021261Z","strongest_claim":"despite reasonable success rates, models frequently exhibit structural hallucinations, missing objects, and failures to satisfy geometric constraints, with limited ability to exploit visual and constraint-based feedback for self-correction","one_line_summary":"GeoBuildBench is a new benchmark requiring LLMs to generate executable geometry constructions from text, revealing frequent hallucinations, missing objects, and constraint failures in state-of-the-art models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 489 problems are text-complete and constructible after automated filtering and human validation, and the chosen DSL accurately represents valid geometric constructions.","pith_extraction_headline":"Current multimodal models often produce geometry diagrams with hallucinations, missing objects, and violated constraints from natural language problems."},"references":{"count":12,"sample":[{"doi":"","year":2021,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","ref_index":1,"cited_arxiv_id":"2511.21631","is_internal_anchor":true},{"doi":"","year":null,"title":"One definition per label","work_id":"65c86816-fcc8-4cc5-8fce-057339080c40","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Construct, don't assert","work_id":"7d863cb8-fce3-4358-9c98-afe4a8cad918","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"These prompts define the construction suitability criteria and the formal extraction of geometric objects and verifica- tion conditions","work_id":"d40e34e9-4394-48d5-b994-6b0e6692437e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Determine whether this problem is suitable for geometric figure construction","work_id":"40f491f4-77ea-48ae-b724-05e0380c7495","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":12,"snapshot_sha256":"eb2354cb310d94239aa3c1c1790363cda58565c67dd64946d2598d9d3058403c","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"f33771f89a64f5d72cbd7ec57900d483471b62b6220cbc15eb76bf15efeb6c21"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}