ARC Prize reports that Claude Fable 5.1 scores 90.0% on ARC-AGI-2 Semi-Private at maximum reasoning effort, with a reported cost of $4.49 per task. Its ARC-AGI-1 Semi-Private result at the same setting is 97.5%, at $1.40 per task.
The results page distinguishes five reasoning settings. On ARC-AGI-2, the listed score is 78.3% at low effort, 86.3% at medium, 88.8% at high and 90.0% at both xhigh and max.
The page also offers task-by-task outcomes for the public evaluation sets. Those tables are separate from the semi-private headline results, and it lists no ARC-AGI-3 score for this model. The reported dollar figures describe evaluation-task costs, not a subscription price or the expected cost of an arbitrary coding task.