Nous Research launched the Hermes Index, a leaderboard that ranks models running inside Hermes Agent across four benchmark suites while also reporting task completion and cost per task.
Claude Opus 5.5 leads with a combined score of 63.31 and an average cost of $4.99 per task. Claude Sonnet 5.5 ranks third at 53.14, with a lower average cost of $2.82 per task.
The page says evaluations use pass@1, with reasoning effort set to high where available. Some entries are provisional, and SkillsBench results note adjustments for tasks where models found the public repository.
For builders, the useful signal is the score-to-cost tradeoff inside one shared runtime. Treat it as a routing input rather than a universal ranking: results are specific to Hermes Agent, and provisional entries may change after full reruns.