ARCBench S1 - c2 Hermes Compacted: leaderboard
ARCBench pi agent harness; same exact endpoint is agent and compactor, 17 task cases with only configurations having all valid repetitions included. Violation@k is percent of cases with at least one violation in k runs; lower is better. GPT-Luna retains the recorded mutable ~openai/gpt-luna-latest endpoint without an inferred dated identity. S1; k=3, c2, Hermes Compacted; no attack or defense.
Metric: Violation@3 (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gpt-luna-latest (ARCBench Run: endpoint=~openai%2Fgpt-luna-latest; harness=pi; context=Hermes%20Compacted; channel=c2) | 28.6 | |
| 2 | deepseek-v4-flash-0731 (ARCBench Run: endpoint=deepseek%2Fdeepseek-v4-flash-0731; harness=pi; context=Hermes%20Compacted; channel=c2) | 37.5 | |
| 3 | glm-5.3-flash (ARCBench Run: endpoint=z-ai%2Fglm-5.3-flash; harness=pi; context=Hermes%20Compacted; channel=c2) | 50 | |
| 4 | qwen3.7-flash (ARCBench Run: endpoint=qwen%2Fqwen3.7-flash; harness=pi; context=Hermes%20Compacted; channel=c2) | 62.5 |
Interactive version: theaggregate.ai/benchmark?slug=arcbench-s1-c2-hermes-compacted · How It Works · Data refreshed daily, snapshot 2026-11-05.