AARRI-Bench (Hermes Agent) - Context: leaderboard

Metric: Task success rate (%) under the 0/1 reward on the context tasks: sensitivity to the state of a field (core contributions, data validity, genuine advances); Harbor containerized runs, Hermes Agent agent harness; a configuration's tasks without a valid run are left out of its denominator; higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 6 models tracked.

Top models

#ModelScore
1DeepSeek V4 Flash55.9
2Claude Opus 4.752.9
3Qwen 3.6 Plus50
4Claude Sonnet 4.647.1
5MiniMax-M2.744.1
6Qwen 3 235B A22B 2507 (Thinking)34.3

Interactive version: theaggregate.ai/benchmark?slug=aarri-bench-hermes-agent-context · How It Works · Data refreshed daily, snapshot 2026-09-29.