{"record": {"aliases": ["VISTA: An End-to-End Benchmark for Visual Spec-to-Web-App Coding Agents"], "artifacts": [{"id": "arxiv:2605.26144", "kind": "paper", "url": "https://arxiv.org/abs/2605.26144"}, {"id": "gh:kaboider/vista_bench", "kind": "repo", "url": "https://github.com/kaboider/VISTA_Bench"}, {"kind": "website", "url": "https://kaboider.github.io/VIS_APP/"}], "authors": ["JunJia Guo", "Yuhang Yao", "Chang Li", "Zimeng Pan", "Jiawei", "Zhou", "Jingdi Chen"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.26144", "source_url": "https://arxiv.org/abs/2605.26144"}, "categories": ["Code & Software", "Software & AI Compute", "cs.SE"], "description": {"en": "A benchmark for LLM coding agents that build web apps from visual specs. It covers five input conditions and scores structure, behavior, and visual fidelity."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.26144", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.26144", "title": "VISTA (VIsual Spec-To-App Benchmark)"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.26144", "modality": null, "name": "VISTA (VIsual Spec-To-App Benchmark)", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_vista-an-end-to-end-benchmark-for-visual-s_68d41fbf", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "04d7889e8406d3381ecc3e72e63507eef9f0d12c3095560e38f4456890b764ec", "relation": "introduces", "review_model": "claude-code-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:46:40.642301Z", "source_url": "https://arxiv.org/abs/2605.26144"}, "publisher": null, "released": "2026-05-22", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.26144", "source_url": "https://arxiv.org/abs/2605.26144"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-26144", "source": "claire_radar", "source_benchmark_id": "2605.26144", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["VISTA: An End-to-End Benchmark for Visual Spec-to-Web-App Coding Agents"], "annotation": "Unknown", "applicationDomains": ["Software & AI Compute"], "area": "Code & Software", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Agentic web-app coding is an active area, and code plus a project page are public.", "score": 45}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:46:40.642301Z", "inputHash": "82f37e716424805090449a9cc4864bae28c23e68bd44e353dabc2b8719970635", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "We present VISTA (VIsual Spec-To-App Benchmark), a benchmark", "canonicalNameSource": "abstract", "decisionReason": "The benchmark is formally named, defines tasks across five prompt conditions, and specifies a fixed composite scoring method. The code is public.", "evaluationEvidence": {"reuse": "Code is available at https://github.com/kaboider/VISTA_Bench.", "scoring": "Evaluation combines DOM-grounded reference matching, behavior-specific browser tests, and CLIP-based visual similarity", "task": "a benchmark for evaluating the end-to-end web-app generation capabilities of LLM-based agents"}, "model": "claude-code-default", "reviewedAt": "2026-09-24T16:46:40.642301Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A benchmark for LLM coding agents that build web apps from visual specs. It covers five input conditions and scores structure, behavior, and visual fidelity.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We present VISTA (VIsual Spec-To-App Benchmark), a benchmark for evaluating the end-to-end web-app generation capabilities of LLM-based agents."}, "familyId": "bmf_2529a397e5cd", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_vista-an-end-to-end-benchmark-for-visual-s_68d41fbf", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/kaboider/VISTA_Bench", "data": null, "paper": "https://arxiv.org/abs/2605.26144", "pdf": "https://arxiv.org/pdf/2605.26144", "project": "https://kaboider.github.io/VIS_APP/", "report": "https://arxiv.org/abs/2605.26144"}, "metrics": [], "motivation": "We present VISTA (VIsual Spec-To-App Benchmark), a benchmark for evaluating the end-to-end web-app generation capabilities of LLM-based agents.", "name": "VISTA (VIsual Spec-To-App Benchmark)", "oneLine": "A benchmark for LLM coding agents that build web apps from visual specs. It covers five input conditions and scores structure, behavior, and visual fidelity.", "paperTitle": "VISTA: An End-to-End Benchmark for Visual Spec-to-Web-App Coding Agents", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.26144", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-05-22", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["JunJia Guo", "Yuhang Yao", "Jiawei (Joe) Zhou", "Jingdi Chen"], "categories": ["cs.SE", "cs.AI", "cs.CV"], "id": "2605.26144", "title": "VISTA: An End-to-End Benchmark for Visual Spec-to-Web-App Coding Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2605.26144"}, "sourceUpdatedAt": "2026-06-22T18:12:46Z", "topics": ["cs.SE"], "venueAttempts": [], "whyItMatters": "It extends code-generation evaluation from algorithmic tasks to realistic UI-centric app building. Manually annotated pages support reproducible checks of both function and visual fidelity."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
