VOYONS MODELS ARENA
Live rankings tracking Voyons model performance. Every forecast is measured. Every comparison is transparent.
OFFICIAL RANKINGS
| RANK | MODEL |
|---|---|
| 1 |
voyons-nano-25.6-release
Rating: 1517
Win Rate: 54%
Released: Jun 01, 2025
|
| 2 |
voyons-tiny-25.7-early
Rating: 1505
Win Rate: 51%
Released: Jul 01, 2025
|
| 3 |
voyons-nano-25.7-release
Rating: 1502
Win Rate: 50%
Released: Jul 01, 2025
|
| 4 |
voyons-tiny-25.6-release
Rating: 1493
Win Rate: 48%
Released: Jun 01, 2025
|
| 5 |
voyons-tiny-25.7-release
Rating: 1489
Win Rate: 47%
Released: Jul 01, 2025
|
| 6 |
voyons-tiny-25.5-release
Rating: 1441
Win Rate: 35%
Released: May 01, 2025
|
HOW WE MEASURE PERFORMANCE
Duon Labs maintains transparent benchmarks for all Voyons models in production. Models are continuously evaluated by forecasting top cryptocurrency pairs across multiple timeframes (15m, 1h, 1d) in real-time market conditions. Each variant competes head-to-head on identical forecasting tasks to determine relative performance.
CONTINUOUS MARKET EVALUATION
Real-time forecasting across crypto markets.
Models continuously forecast major cryptocurrency pairs (BTC/USDT, ETH/USDT, and more) across multiple timeframes: 15-minute for rapid movements, 1-hour for intraday patterns, and daily for longer trends. This diverse evaluation ensures models perform well across different market conditions and trading horizons.
timeframes: [15m, 1h, 1d] × top_crypto_pairs
EXTROPY-BASED SCORING
The metric that matters: forecast precision.
Each model's forecast quality is measured using extropy, an information-theoretic metric. We evaluate precision by using the input candle data to forecast the next 3 candles, then calculate extropy between the actual samples and forecasted samples. This measures how well the model captures the underlying patterns in the data.
response_last_candle_extropy = AVG(last_3_candles_extropies)
GLICKO-2 RATING SYSTEM
Advanced skill-based rankings.
We use the Glicko-2 rating system to rank models based on head-to-head performance. Each model starts at 1500 rating points. Winning against stronger opponents provides more rating gain, while losing to weaker opponents results in larger rating drops. The system converges quickly to each model's true skill level.
starting_rating: 1500 | higher_is_better
QUALIFICATION CRITERIA
Statistical significance before ranking.
To ensure meaningful rankings, models must prove themselves through volume and diversity: minimum 100 games played against at least 2 different opponents. This ensures that rankings reflect consistent performance across varied competition rather than isolated results against a single opponent.
qualified: games ≥ 100 AND opponents ≥ 2
EXPLORE VOYONS
Experience the power of Voyons' probabilistic forecasting. Access these models through our API and build your own applications.