{
  "schemaVersion": 1,
  "kind": "benchmark-index",
  "count": 12,
  "generatedAt": "2026-09-10T00:00:00.000Z",
  "scoreNote": "This index lists what each benchmark measures, how it is scored, and the dates its sources were checked. It does not include model score rows — cite the sourced ledger on each benchmark page.",
  "benchmarks": [
    {
      "slug": "aa-lcr",
      "name": "AA-LCR",
      "family": "long-context",
      "status": "solid",
      "firstParty": false,
      "summary": {
        "en": "Long-context synthesis and reasoning: 100 open-answer questions requiring models to integrate evidence across long inputs.",
        "de": "Long-Context-Synthese und Reasoning: 100 offene Antwortfragen, die Modelle zwingen, Evidenz über lange Inputs zu integrieren."
      },
      "taskFormat": {
        "en": "100 open-answer questions with long context passages; equality-checker LLM grading, pass@1.",
        "de": "100 offene Antwortfragen mit langen Kontextpassagen; Equality-Checker-LLM-Bewertung, pass@1."
      },
      "scoringMethod": {
        "en": "pass@1 accuracy with 3 repeats.",
        "de": "pass@1-Genauigkeit mit 3 Wiederholungen."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Artificial Analysis",
      "releaseYear": 2025,
      "sources": [
        {
          "label": {
            "en": "AA-LCR evaluation",
            "de": "AA-LCR-Evaluation"
          },
          "url": "https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning",
          "checkedAt": "2026-06-22"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/aa-lcr"
    },
    {
      "slug": "aa-omniscience",
      "name": "AA-Omniscience",
      "family": "knowledge",
      "status": "flagship",
      "firstParty": false,
      "summary": {
        "en": "Factual recall and knowledge calibration across 6,000 questions in 42 economically relevant topics and six domains. The score splits accuracy from a non-hallucination slice.",
        "de": "Faktenabruf und Wissenskalibrierung über 6.000 Fragen in 42 wirtschaftlich relevanten Themen und sechs Domänen. Der Score trennt Accuracy von einem Non-Hallucination-Anteil."
      },
      "taskFormat": {
        "en": "6,000 questions from academic and industry sources across business, humanities, science/engineering, health, law, and software engineering. Models may answer or abstain.",
        "de": "6.000 Fragen aus akademischen und Branchenquellen über Business, Geisteswissenschaften, Science/Engineering, Gesundheit, Recht und Software Engineering. Modelle dürfen antworten oder sich enthalten."
      },
      "scoringMethod": {
        "en": "VerdictPal ledger rows are AA-Omniscience Accuracy (% of all questions answered correctly, including abstentions as misses). The companion Index (−100 to +100) rewards correct answers, penalizes hallucinations, and scores 0 for abstention — we keep Index figures in notes, not as the sortable score, because the Index can go negative.",
        "de": "VerdictPal-Ledger-Zeilen sind AA-Omniscience-Accuracy (% aller Fragen korrekt, Enthaltungen zählen als Fehler). Der Begleit-Index (−100 bis +100) belohnt korrekte Antworten, bestraft Hallucinationen und zählt 0 für Enthaltung — Index-Zahlen stehen in den Notes, nicht als sortierbarer Score, weil der Index negativ werden kann."
      },
      "contaminationRisk": "medium",
      "saturation": "active",
      "maintainer": "Artificial Analysis",
      "releaseYear": 2025,
      "sources": [
        {
          "label": {
            "en": "AA-Omniscience evaluation",
            "de": "AA-Omniscience-Evaluation"
          },
          "url": "https://artificialanalysis.ai/evaluations/omniscience",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "AA-Omniscience paper (arXiv 2511.13029)",
            "de": "AA-Omniscience-Paper (arXiv 2511.13029)"
          },
          "url": "https://arxiv.org/abs/2511.13029",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "Public question subset (Hugging Face)",
            "de": "Öffentliches Fragen-Subset (Hugging Face)"
          },
          "url": "https://huggingface.co/datasets/ArtificialAnalysis/AA-Omniscience-Public",
          "checkedAt": "2026-09-01"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/aa-omniscience"
    },
    {
      "slug": "aa-output-speed",
      "name": "AA output speed",
      "family": "performance",
      "status": "flagship",
      "firstParty": false,
      "summary": {
        "en": "Median output tokens per second after the first chunk arrives, measured on each model's default API provider with a ~1k-token input prompt (AA medium workload).",
        "de": "Medianer Output-Tokens pro Sekunde nach dem ersten Chunk, gemessen am Default-API-Provider jedes Modells mit ~1k Input-Tokens (AA-Medium-Workload)."
      },
      "taskFormat": {
        "en": "Streaming generation after first token; median over the past 72 hours on AA's harness.",
        "de": "Streaming-Generierung nach erstem Token; Median der letzten 72 Stunden im AA-Harness."
      },
      "scoringMethod": {
        "en": "Higher tokens/s is faster. AA uses OpenAI tiktoken o200k_base for token counts.",
        "de": "Mehr Tokens/s ist schneller. AA nutzt OpenAI tiktoken o200k_base für Token-Zählung."
      },
      "contaminationRisk": "unknown",
      "saturation": "active",
      "maintainer": "Artificial Analysis",
      "releaseYear": 2026,
      "sources": [
        {
          "label": {
            "en": "AA performance methodology",
            "de": "AA-Performance-Methodik"
          },
          "url": "https://artificialanalysis.ai/methodology/performance-benchmarking",
          "checkedAt": "2026-06-10"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/aa-output-speed"
    },
    {
      "slug": "critpt",
      "name": "CritPt",
      "family": "reasoning",
      "status": "solid",
      "firstParty": false,
      "summary": {
        "en": "Unpublished research-level physics: 71 composite challenges (70 test + 1 example) written by 50+ active physicists across 11 subfields, plus 190 simpler checkpoint tasks.",
        "de": "Unveröffentlichte Forschungsphysik: 71 zusammengesetzte Challenges (70 Test + 1 Beispiel) von 50+ aktiven Physiker:innen über 11 Teilgebiete, plus 190 einfachere Checkpoint-Aufgaben."
      },
      "taskFormat": {
        "en": "71 composite research challenges spanning condensed matter, quantum, AMO, astrophysics, HEP, mathematical physics, statistical physics, nuclear, nonlinear dynamics, fluids, and biophysics. Models may use coding tools in some AA runs; checkpoint tasks (190) are a separate, easier slice.",
        "de": "71 zusammengesetzte Research-Challenges über Festkörper, Quanten, AMO, Astrophysik, HEP, mathematische Physik, Statistik, Kernphysik, nichtlineare Dynamik, Fluide und Biophysik. Modelle dürfen in manchen AA-Läufen Coding-Tools nutzen; Checkpoint-Aufgaben (190) sind ein separates, leichteres Slice."
      },
      "scoringMethod": {
        "en": "Average accuracy on the 70-test composite set. Automated grading for physics-specific output formats. VerdictPal rows use Artificial Analysis' independent CritPt run.",
        "de": "Durchschnittsgenauigkeit auf dem 70-Test-Composite-Set. Automatische Bewertung für physikspezifische Ausgabeformate. VerdictPal-Zeilen nutzen Artificial Analysis' unabhängigen CritPt-Lauf."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Argonne / UIUC / Artificial Analysis",
      "releaseYear": 2025,
      "sources": [
        {
          "label": {
            "en": "AA CritPt evaluation",
            "de": "AA-CritPt-Evaluation"
          },
          "url": "https://artificialanalysis.ai/evaluations/critpt",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "CritPt paper (arXiv 2509.26574)",
            "de": "CritPt-Paper (arXiv 2509.26574)"
          },
          "url": "https://arxiv.org/abs/2509.26574",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "CritPt official site",
            "de": "CritPt-Offizielle Seite"
          },
          "url": "https://critpt.com/",
          "checkedAt": "2026-09-01"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/critpt"
    },
    {
      "slug": "deepswe",
      "name": "DeepSWE",
      "family": "coding",
      "status": "flagship",
      "firstParty": false,
      "summary": {
        "en": "Long-horizon software engineering on 113 contamination-free tasks across 91 repos and five languages — patches must pass hand-written behavioral verifiers, not string diffs.",
        "de": "Langhorizont-Software-Engineering auf 113 kontaminationsfreien Aufgaben über 91 Repos und fünf Sprachen — Patches müssen handgeschriebene Verhaltens-Verifier bestehen, keine String-Diffs."
      },
      "taskFormat": {
        "en": "113 tasks from scratch (not mined PRs); pass@1 via behavioral verifiers; mini-swe-agent harness.",
        "de": "113 Aufgaben from scratch (keine geminten PRs); pass@1 über Verhaltens-Verifier; mini-swe-agent-Harness."
      },
      "scoringMethod": {
        "en": "pass@1 (% resolved).",
        "de": "pass@1 (% gelöst)."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Datacurve",
      "releaseYear": 2026,
      "sources": [
        {
          "label": {
            "en": "DeepSWE leaderboard",
            "de": "DeepSWE-Leaderboard"
          },
          "url": "https://deepswe.datacurve.ai/",
          "checkedAt": "2026-06-22"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/deepswe"
    },
    {
      "slug": "frontiermath",
      "name": "FrontierMath",
      "family": "math",
      "status": "flagship",
      "firstParty": false,
      "summary": {
        "en": "Hard unpublished mathematics: 295 private Tiers 1-3 problems (v2) where the model writes a Python answer() after iterative reasoning and code execution.",
        "de": "Harte unveröffentlichte Mathematik: 295 private Tiers-1-3-Aufgaben (v2), bei denen das Modell nach iterativem Reasoning und Code-Ausführung eine Python-answer()-Funktion abgibt."
      },
      "taskFormat": {
        "en": "Private set of 295 Tiers 1-3 problems (v2, 12 Jun 2026 correction). The model may think aloud, call a stateless Python tool, and must submit an answer() function. Token cap 1,000,000.",
        "de": "Privates Set aus 295 Tiers-1-3-Aufgaben (v2, Korrektur 12. Jun. 2026). Das Modell darf laut denken, ein zustandsloses Python-Tool aufrufen und muss eine answer()-Funktion abgeben. Token-Limit 1.000.000."
      },
      "scoringMethod": {
        "en": "Pass rate on the private Tiers 1-3 v2 set (1 if the submitted object matches, else 0). VerdictPal rows use Epoch's best internal run per model from the public CSV.",
        "de": "Pass-Rate auf dem privaten Tiers-1-3-v2-Set (1 wenn das abgegebene Objekt stimmt, sonst 0). VerdictPal-Zeilen nutzen Epochs besten internen Lauf pro Modell aus der öffentlichen CSV."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Epoch AI",
      "releaseYear": 2024,
      "sources": [
        {
          "label": {
            "en": "Epoch FrontierMath Tiers 1-3 (v2)",
            "de": "Epoch FrontierMath Tiers 1-3 (v2)"
          },
          "url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2",
          "checkedAt": "2026-08-15"
        },
        {
          "label": {
            "en": "Epoch benchmarking data download",
            "de": "Epoch-Benchmarking-Daten-Download"
          },
          "url": "https://epoch.ai/benchmarks/use-this-data",
          "checkedAt": "2026-08-15"
        },
        {
          "label": {
            "en": "FrontierMath program",
            "de": "FrontierMath-Programm"
          },
          "url": "https://epoch.ai/frontiermath",
          "checkedAt": "2026-08-15"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/frontiermath"
    },
    {
      "slug": "gpqa-diamond",
      "name": "GPQA Diamond",
      "family": "reasoning",
      "status": "solid",
      "firstParty": false,
      "summary": {
        "en": "GPQA Diamond explained: graduate-level Google-proof Q&A in biology, physics, and chemistry. Items are written so non-experts fail even with web search; Diamond is the expert-validated hard subset (198 multiple-choice questions). A high score means correct answers on that set under the published protocol — not a general research-assistant grade.",
        "de": "GPQA Diamond erklärt: Google-sichere Q&A auf Graduiertenniveau in Biologie, Physik und Chemie. Die Items sind so geschrieben, dass Laien auch mit Websuche scheitern; Diamond ist die von Experten geprüfte schwere Teilmenge (198 Multiple-Choice-Fragen). Ein hoher Score heisst korrekte Antworten auf diesem Satz nach dem veröffentlichten Protokoll — keine allgemeine Note für Recherche-Assistenten."
      },
      "taskFormat": {
        "en": "198 expert-validated multiple-choice questions in the Diamond set; designed to be 'Google-proof'.",
        "de": "198 von Experten geprüfte Multiple-Choice-Fragen im Diamond-Satz; 'Google-sicher' konzipiert."
      },
      "scoringMethod": {
        "en": "Accuracy (% correct).",
        "de": "Genauigkeit (% korrekt)."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Rein et al.",
      "releaseYear": 2023,
      "sources": [
        {
          "label": {
            "en": "Paper (Rein et al.)",
            "de": "Paper (Rein et al.)"
          },
          "url": "https://arxiv.org/abs/2311.12022",
          "checkedAt": "2026-05-30"
        },
        {
          "label": {
            "en": "Also tracked on Vals AI (independent run)",
            "de": "Auch auf Vals AI (eigener Lauf)"
          },
          "url": "https://www.vals.ai/benchmarks",
          "checkedAt": "2026-06-09"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/gpqa-diamond"
    },
    {
      "slug": "hle",
      "name": "Humanity's Last Exam",
      "family": "reasoning",
      "status": "flagship",
      "firstParty": false,
      "summary": {
        "en": "Cross-domain expert-level questions — math, sciences, humanities, professional law and medicine — designed to be the hardest public exam a frontier model can take.",
        "de": "Fachübergreifende Fragen auf Expertenniveau — Mathematik, Naturwissenschaften, Geisteswissenschaften, professionelles Recht und Medizin — entworfen als die härteste öffentliche Prüfung, die ein Frontier-Modell ablegen kann."
      },
      "taskFormat": {
        "en": "Multiple-choice and short-answer questions vetted by domain experts; submissions are reviewed for difficulty before scoring.",
        "de": "Multiple-Choice- und Kurzantwortfragen, geprüft von Fachexpert:innen; Einreichungen werden vor der Bewertung auf Schwierigkeit geprüft."
      },
      "scoringMethod": {
        "en": "Percent of questions answered correctly on the public + private split; HLE reports a single accuracy figure per model.",
        "de": "Prozent korrekt beantworteter Fragen am öffentlichen + privaten Split; HLE berichtet eine Genauigkeitszahl pro Modell."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Center for AI Safety / Scale AI",
      "releaseYear": 2025,
      "sources": [
        {
          "label": {
            "en": "Humanity's Last Exam project",
            "de": "Humanity's Last Exam Projekt"
          },
          "url": "https://lastexam.ai/",
          "checkedAt": "2026-06-09"
        },
        {
          "label": {
            "en": "Artificial Analysis HLE evaluation",
            "de": "Artificial-Analysis-HLE-Evaluation"
          },
          "url": "https://www.artificialanalysis.ai/evaluations/humanitys-last-exam",
          "checkedAt": "2026-06-09"
        },
        {
          "label": {
            "en": "HLE paper (arXiv)",
            "de": "HLE-Paper (arXiv)"
          },
          "url": "https://arxiv.org/abs/2501.14249",
          "checkedAt": "2026-06-09"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/hle"
    },
    {
      "slug": "mmmu-pro",
      "name": "MMMU-Pro",
      "family": "multimodal",
      "status": "solid",
      "firstParty": false,
      "summary": {
        "en": "Harder multimodal college exam: original MMMU items with text-only shortcuts filtered out, ten answer options instead of four, and a vision-only setting where the question is inside the image. 3,460 questions across 30 subjects.",
        "de": "Härtere multimodale Hochschulprüfung: originale MMMU-Items ohne Text-only-Shortcuts, zehn statt vier Optionen, und ein Vision-only-Setting, in dem die Frage im Bild steckt. 3.460 Fragen über 30 Fächer."
      },
      "taskFormat": {
        "en": "3,460 image+text (and vision-only) items across art, business, science, health, humanities, and engineering. Ten multiple-choice options after filtering questions a text-only model can solve.",
        "de": "3.460 Bild+Text- (und Vision-only-)Items über Kunst, Business, Science, Gesundheit, Geisteswissenschaften und Engineering. Zehn Multiple-Choice-Optionen nach Filterung von Fragen, die ein Text-only-Modell lösen kann."
      },
      "scoringMethod": {
        "en": "Accuracy (% correct). VerdictPal ledger rows are Artificial Analysis' independent MMMU-Pro run.",
        "de": "Genauigkeit (% korrekt). VerdictPal-Ledger-Zeilen sind Artificial Analysis' unabhängiger MMMU-Pro-Lauf."
      },
      "contaminationRisk": "medium",
      "saturation": "active",
      "maintainer": "Yue et al. / Artificial Analysis",
      "releaseYear": 2024,
      "sources": [
        {
          "label": {
            "en": "AA MMMU-Pro evaluation",
            "de": "AA-MMMU-Pro-Evaluation"
          },
          "url": "https://artificialanalysis.ai/evaluations/mmmu-pro",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "MMMU-Pro paper (arXiv 2409.02813)",
            "de": "MMMU-Pro-Paper (arXiv 2409.02813)"
          },
          "url": "https://arxiv.org/abs/2409.02813",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "Hugging Face MMMU_Pro",
            "de": "Hugging-Face-MMMU_Pro"
          },
          "url": "https://huggingface.co/datasets/MMMU/MMMU_Pro",
          "checkedAt": "2026-09-01"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/mmmu-pro"
    },
    {
      "slug": "scicode",
      "name": "SciCode",
      "family": "coding",
      "status": "solid",
      "firstParty": false,
      "summary": {
        "en": "Scientific programming tasks requiring domain libraries and multi-step numerical code — AA snapshot runs.",
        "de": "Wissenschaftliche Programmieraufgaben mit Domänen-Bibliotheken und mehrstufigem numerischem Code — AA-Snapshot-Läufe."
      },
      "taskFormat": {
        "en": "Scientific coding problems with library dependencies.",
        "de": "Wissenschaftliche Coding-Probleme mit Bibliotheksabhängigkeiten."
      },
      "scoringMethod": {
        "en": "Percentage solved from AA runs.",
        "de": "Gelöst-Prozent aus AA-Läufen."
      },
      "contaminationRisk": "medium",
      "saturation": "active",
      "maintainer": "SciCode authors",
      "releaseYear": 2024,
      "sources": [
        {
          "label": {
            "en": "AA SciCode",
            "de": "AA SciCode"
          },
          "url": "https://www.artificialanalysis.ai/evaluations/scicode",
          "checkedAt": "2026-06-10"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/scicode"
    },
    {
      "slug": "terminal-bench-science",
      "name": "Terminal-Bench-Science",
      "family": "agentic",
      "status": "flagship",
      "firstParty": false,
      "summary": {
        "en": "Whether an agent can finish real scientific research workflows in a terminal: 70 expert-curated tasks across life, physical, Earth, mathematical, and engineering sciences, graded on concrete artifacts (analyses, simulations, proofs, code, data products).",
        "de": "Ob ein Agent echte wissenschaftliche Research-Workflows im Terminal zu Ende bringt: 70 von Expert:innen kuratierte Aufgaben über Life, Physical, Earth, Mathematical und Engineering Sciences, bewertet an konkreten Artefakten (Analysen, Simulationen, Beweise, Code, Datenprodukte)."
      },
      "taskFormat": {
        "en": "70 containerized Harbor tasks. Agents get a terminal, files, and tools; graders check the produced artifact with task-specific tests. Three independent trials per task in the 0.1 announcement table.",
        "de": "70 containerisierte Harbor-Aufgaben. Agenten bekommen Terminal, Dateien und Tools; Grader prüfen das Artefakt mit aufgabenspezifischen Tests. Drei unabhängige Trials pro Aufgabe in der 0.1-Announcement-Tabelle."
      },
      "scoringMethod": {
        "en": "Resolution rate across 70 tasks (pass if the hidden checker accepts the artifact). VerdictPal rows use the 27 Aug 2026 announcement table, one row per model at its published harness.",
        "de": "Resolution-Rate über 70 Aufgaben (bestanden, wenn der versteckte Checker das Artefakt akzeptiert). VerdictPal-Zeilen nutzen die Announcement-Tabelle vom 27. Aug. 2026, eine Zeile pro Modell mit dem veröffentlichten Harness."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Stanford / Laude Institute / Terminal-Bench",
      "releaseYear": 2026,
      "sources": [
        {
          "label": {
            "en": "Terminal-Bench-Science 0.1 announcement",
            "de": "Terminal-Bench-Science-0.1-Announcement"
          },
          "url": "https://www.tbench.ai/news/terminal-bench-science-0-1",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "Live leaderboard",
            "de": "Live-Leaderboard"
          },
          "url": "https://www.terminal-bench-science.ai/",
          "checkedAt": "2026-09-01"
        },
        {
          "label": {
            "en": "GitHub task suite",
            "de": "GitHub-Aufgabensuite"
          },
          "url": "https://github.com/harbor-framework/terminal-bench-science",
          "checkedAt": "2026-09-01"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/terminal-bench-science"
    },
    {
      "slug": "terminal-bench",
      "name": "Terminal-Bench",
      "family": "agentic",
      "status": "solid",
      "firstParty": false,
      "summary": {
        "en": "Whether an agent can complete realistic, containerized terminal tasks (Terminal-Bench 2.1 on the public AA API): debugging code, handling files, running commands, and producing the artifact the hidden checker expects. Index v4.2 weighted this suite at 10%; Index v4.3 uses Terminal-Bench v4.0 in the composite instead.",
        "de": "Ob ein Agent realistische, containerisierte Terminal-Aufgaben (Terminal-Bench 2.1 in der öffentlichen AA-API) erledigen kann: Code debuggen, Dateien bearbeiten, Befehle ausführen und das Artefakt liefern, das der versteckte Checker erwartet. Index v4.2 gewichtete diese Suite mit 10 %; Index v4.3 nutzt im Composite Terminal-Bench v4.0."
      },
      "taskFormat": {
        "en": "Containerized tasks with a terminal environment; agents operate through commands and are graded by task-specific checks.",
        "de": "Containerisierte Aufgaben mit Terminal-Umgebung; Agenten arbeiten über Befehle und werden durch aufgabenspezifische Checks bewertet."
      },
      "scoringMethod": {
        "en": "Resolution/pass rate across audited tasks. Always read the agent harness and model pairing.",
        "de": "Lösungs-/Passrate über auditierte Aufgaben. Agent-Harness und Modell-Paarung immer mitlesen."
      },
      "contaminationRisk": "low",
      "saturation": "active",
      "maintainer": "Laude Institute / Terminal-Bench",
      "releaseYear": 2025,
      "sources": [
        {
          "label": {
            "en": "Official Terminal-Bench leaderboard",
            "de": "Offizielles Terminal-Bench-Leaderboard"
          },
          "url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0",
          "checkedAt": "2026-09-10"
        },
        {
          "label": {
            "en": "Terminal-Bench 2 repository",
            "de": "Terminal-Bench-2-Repository"
          },
          "url": "https://github.com/laude-institute/terminal-bench-2",
          "checkedAt": "2026-09-10"
        },
        {
          "label": {
            "en": "Artificial Analysis Terminal-Bench Hard",
            "de": "Artificial Analysis Terminal-Bench Hard"
          },
          "url": "https://artificialanalysis.ai/evaluations/terminalbench-hard",
          "checkedAt": "2026-09-10"
        }
      ],
      "html": "https://verdictpal.com/benchmarks/terminal-bench"
    }
  ]
}