Leaderboard

Model evaluation leaderboard and history endpoints

Get Leaderboard

Retrieve the model leaderboard with Elo ratings based on user feedback comparisons.

The leaderboard uses an Elo rating system where:

• Each model starts with a rating of 1000
• Ratings are adjusted based on head-to-head comparisons from user feedback
• The K-factor of 32 controls rating volatility
• Optional query-based filtering uses semantic similarity to weight relevant feedbacks

Query Parameters

query · string

Filter leaderboard by tag similarity. When provided, uses semantic embeddings to compute relevance scores and weight the Elo calculations. This creates topic-specific leaderboards (e.g., “coding” shows which models perform best for coding tasks).

Response

entries · array

Array of leaderboard entries sorted by rating (highest first)

Show LeaderboardEntry properties

model_id · string

Unique identifier for the model

rating · integer

Elo rating rounded to nearest integer (base rating: 1000)

won · integer

Number of comparisons won against other models

lost · integer

Number of comparisons lost against other models

count · integer

Total number of comparisons (won + lost)

top_tags · array

Most frequent tags associated with this model’s feedback

Show Tag properties

tag · string

Tag name

count · integer

Number of times this tag appears

cURL

curl -X GET "https://your-instance.com/api/evaluations/leaderboard" \
  -H "Authorization: Bearer $OPENWEBUI_API_KEY"

{
  "entries": [
    {
      "model_id": "gpt-4",
      "rating": 1245,
      "won": 42,
      "lost": 18,
      "count": 60,
      "top_tags": [
        {"tag": "coding", "count": 25},
        {"tag": "reasoning", "count": 18},
        {"tag": "creative", "count": 12}
      ]
    },
    {
      "model_id": "claude-3",
      "rating": 1198,
      "won": 38,
      "lost": 22,
      "count": 60,
      "top_tags": [
        {"tag": "writing", "count": 30},
        {"tag": "analysis", "count": 15}
      ]
    }
  ]
}

---

Get Model History

Retrieve the daily win/loss history for a specific model over a time period.

Path Parameters

model_id · string · required

Unique identifier for the model

Query Parameters

days · integer · default: "30"

Number of days of history to retrieve (default: 30)

Response

model_id · string

Unique identifier for the model

history · array

Array of daily win/loss records

Show ModelHistoryEntry properties

date · string

Date in ISO 8601 format (YYYY-MM-DD)

won · integer

Number of comparisons won on this date

lost · integer

Number of comparisons lost on this date

cURL

curl -X GET "https://your-instance.com/api/evaluations/leaderboard/gpt-4/history" \
  -H "Authorization: Bearer $OPENWEBUI_API_KEY"

{
  "model_id": "gpt-4",
  "history": [
    {
      "date": "2026-03-02",
      "won": 5,
      "lost": 2
    },
    {
      "date": "2026-03-01",
      "won": 8,
      "lost": 3
    },
    {
      "date": "2026-02-28",
      "won": 6,
      "lost": 4
    }
  ]
}