This page explains the documented scope and methodology of the AlphaBench benchmark. It is educational information, not betting advice.
Models included
The benchmark includes 8 active models:
- deepseek-v4.1-flash
- claude-haiku-5.5
- gpt-6-luna-pro
- apodex-1.1-mini
- nemotron-3.5-lightning
- ling-3.1-flash
- solar-mini4
- qwen3.8-omni-flash
See model evaluation for related information.
Fixture selection
The pilot covers 5 leagues, 20 teams, and 100 fixtures:
- Premier League
- LaLiga
- Bundesliga
- Serie A
- Champions League
This describes the pilot scope, not a claim of coverage beyond it.
Data available to each model
The documented data sources are fixture API import, odds snapshots, and a fixture news store. The available information does not specify the exact fields or timing of access for each model. See data sources.
Odds snapshots and timing
Odds snapshots are among the benchmark’s data sources. The documented methodology does not specify snapshot frequency or exact capture times. Pre-kickoff prediction snapshots are immutable; post-kickoff edits are rejected.
Probability collection and bet selection
The documented facts do not specify how model probabilities are collected, combined, or used to select bets. No additional probability-collection or selection rules are asserted here. This benchmark is not betting advice.
Evaluation metrics
The benchmark records 2,660 settled bets and 1 public correction. These are reported counts; the documented facts do not provide further metric definitions or performance results. See settlement rules.
Exclusions and limitations
The documented scope is the five-league, 20-team, 100-fixture pilot. The available methodology does not establish coverage outside that scope or specify additional data, timing, probability-collection, or selection procedures. Pre-kickoff prediction snapshots are immutable, and post-kickoff edits are rejected.