- Source value
- 68.8
- Metric
- accuracy
- Unit
- %
- Benchmark version
- Fastino API alias glide, ProofWriter rerun 2026-10-03
- Category
- decision-model-benchmark
- Direction
- Higher is better
- Sample
- 1800
- Source type
- vendor-reported
- Evaluator
- Ryan Porter, Anthus, Glide Decision Model evaluation
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Vendor-reported result. ProofWriter closed-world split, one answer per problem, Fastino API alias “glide”. The 3 October rerun was requested by Fastino's CEO and used the same request as the earlier run. N=1800. Evaluator reports 1,800 open-world and 1,800 closed-world problems.
- Context
- Anthus reports a Jev-related evaluation and may have commercial or competitive interests, so this is conservatively classified vendor-reported. No model revision or full decoding configuration is pinned. The two runs are kept separate, not averaged; rerun AUROC values are not published.