SCOPE - AFlow Escalation: leaderboard
SCOPE - AFlow Escalation. The SCOPE paper compares GPT-5.4 nano, GPT-5.4 mini, and Gemini 3.5 Flash with reasoning disabled on fixed test splits. Each prompt or generated-program condition is kept separate. Optimized conditions use Claude Opus 4.8 as the common analyzer, decomposer, adversarial helper, or teacher; programs remain specific to their executor. Results average nine measurements: three optimization runs and three test repetitions. Direct prompting uses the original task specification. Coaching, Escalation, and Plan report classification accuracy. Tile-match reports mean Jaccard overlap between predicted and correct label sets, not exact-match accuracy. Reported metric: Accuracy (0–1).
Metric: Accuracy (0–1). Source: arxiv.org. Saturation forecast: Around January 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 Mini (Non-reasoning) | 0.92 | #216 (GPT-5.4 Mini) |
| 2 | GPT-5.4 Nano (Non-reasoning) | 0.86 | #348 (GPT-5.4 Nano) |
No result here: #5 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scope-aflow-escalation · How It Works · Data refreshed daily, snapshot 2026-10-25.