OverviewStructure & Data › R44
Structure & Data · blind round

R44 · Appointment extraction to strict JSON

blind round 8/8 local models answered task written in German

Answers below are anonymized and shuffled. Read, judge, pick a favourite — then scroll down and reveal the models.

Show task
Extrahiere alle Termine aus folgendem Text als JSON-Array von Objekten mit exakt den Feldern datum (YYYY-MM-DD), uhrzeit (HH:MM, 24h) und titel (String). Gib NUR das JSON aus, ohne Erklärung.

Text: "Am Dienstag, den 12. August 2026, treffen wir uns um halb drei nachmittags zur Budgetplanung. Der Zahnarzttermin ist am 03.09.2026 um 9 Uhr morgens. Nicht vergessen: Silvesterparty am 31.12. um 20:30! Das Sommerfest wurde übrigens abgesagt."

The answers — blind

Local models (Brain cluster) — identical prompt, anonymized order
MODEL Amistral-small:24b
```json
[
  {"datum": "2026-08-12", "uhrzeit": "15:30", "titel": "Budgetplanung"},
  {"datum": "2026-09-03", "uhrzeit": "09:00", "titel": "Zahnarzttermin"},
  {"datum": "2026-12-31", "uhrzeit": "20:30", "titel": "Silvesterparty"}
]
```
MODEL Bqwen3-coder:30b
```json
[
  {
    "datum": "2026-08-12",
    "uhrzeit": "14:30",
    "titel": "Budgetplanung"
  },
  {
    "datum": "2026-09-03",
    "uhrzeit": "09:00",
    "titel": "Zahnarzttermin"
  },
  {
    "datum": "2026-12-31",
    "uhrzeit": "20:30",
    "titel": "Silvesterparty"
  }
]
```
MODEL Cgemma4:26b
```json
[
  {
    "datum": "2026-08-12",
    "uhrzeit": "14:30",
    "titel": "Budgetplanung"
  },
  {
    "datum": "2026-09-03",
    "uhrzeit": "09:00",
    "titel": "Zahnarzttermin"
  },
  {
    "datum": "2026-12-31",
    "uhrzeit": "20:30",
    "titel": "Silvesterparty"
  }
]
```
MODEL Dqwen3:8b
[
  {
    "datum": "2026-08-12",
    "uhrzeit": "15:30",
    "titel": "Budgetplanung"
  },
  {
    "datum": "2026-09-03",
    "uhrzeit": "09:00",
    "titel": "Zahnarzttermin"
  },
  {
    "datum": "2026-12-31",
    "uhrzeit": "20:30",
    "titel": "Silvesterparty"
  }
]
MODEL Edeepseek-r1:14b
[
  {
    "datum": "2026-08-12",
    "uhrzeit": "15:30",
    "titel": "Budgetplanung"
  },
  {
    "datum": "2026-09-03",
    "uhrzeit": "09:00",
    "titel": "Zahnarzttermin"
  },
  {
    "datum": "2026-12-31",
    "uhrzeit": "20:30",
    "titel": "Silvesterparty"
  }
]
MODEL Fqwen3:14b
```json
[
  {
    "datum": "2026-08-12",
    "uhrzeit": "14:30",
    "titel": "Budgetplanung"
  },
  {
    "datum": "2026-09-03",
    "uhrzeit": "09:00",
    "titel": "Zahnarzttermin"
  },
  {
    "datum": "2026-12-31",
    "uhrzeit": "20:30",
    "titel": "Silvesterparty"
  }
]
```
MODEL Gllama3.1:8b
[
    {
        "datum": "2026-08-12",
        "uhrzeit": "15:30",
        "titel": "Budgetplanung"
    },
    {
        "datum": "2026-09-03",
        "uhrzeit": "09:00",
        "titel": "Zahnarzttermin"
    },
    {
        "datum": "2026-12-31",
        "uhrzeit": "20:30",
        "titel": "Silvesterparty"
    }
]
MODEL Hqwen3.5:9b
```json
[
  {
    "datum": "2026-08-12",
    "uhrzeit": "15:30",
    "titel": "Budgetplanung"
  },
  {
    "datum": "2026-09-03",
    "uhrzeit": "09:00",
    "titel": "Zahnarzttermin"
  },
  {
    "datum": "2026-12-31",
    "uhrzeit": "20:30",
    "titel": "Silvesterparty"
  }
]
```
Reference
Gold — the correct answerreference
[{"datum":"2026-08-12","uhrzeit":"14:30","titel":"Budgetplanung"},{"datum":"2026-09-03","uhrzeit":"09:00","titel":"Zahnarzttermin"},{"datum":"2026-12-31","uhrzeit":"20:30","titel":"Silvesterparty"}] — abgesagtes Sommerfest darf nicht erscheinen; 'halb drei nachmittags' = 14:30; Jahr für Silvester aus Kontext 2026.
Reveal the models— compare the answers first, then open. Names, blind scores and latencies are hidden until you do.

Who wrote what

Blind labelModelBlind scoreLatencyStatus
MODEL A mistral-small:24b 8.031.7 sOK
MODEL B qwen3-coder:30b 10.012.3 sOK
MODEL C gemma4:26b 10.0129.9 sOK
MODEL D qwen3:8b 8.012.4 sOK
MODEL E deepseek-r1:14b 8.031.2 sOK
MODEL F qwen3:14b 10.014.4 sOK
MODEL G llama3.1:8b 8.0128.5 sOK
MODEL H qwen3.5:9b 8.012.1 sOK
Verdict

8 of 8 answering local models matched the gold answer (blind score ≥ 8): gemma4:26b, mistral-small:24b, llama3.1:8b, qwen3-coder:30b, qwen3:14b, qwen3.5:9b, qwen3:8b, deepseek-r1:14b.