How we score AI agents.
AgentsAI is independent. We rank agents on merit: every tool, including ones built by companies we know, is scored against the same rubric. Here is exactly how it works.
The five sub-scores
Each agent is rated 0–10 on five criteria. The overall score is the weighted average of the sub-scores.
- 01
Capability
How powerful and complete the agent is at its core job — depth of features, model quality, and how well it handles complex, real-world tasks.
30% - 02
Ease of use
How quickly you can get value — setup, onboarding, UI/UX, and day-to-day ergonomics.
20% - 03
Value for money
What you get for the price across tiers, including the free plan and overage costs.
20% - 04
Reliability
Consistency, uptime, accuracy and how gracefully it handles edge cases.
20% - 05
Support & docs
Quality of documentation, community, and responsiveness of human support.
10%
How the overall score is calculated
We multiply each sub-score by its weight and add them up. When a sub-score doesn't apply to a particular agent, the remaining weights are renormalised, so no tool is penalised for a criterion that isn't relevant to it. Scores are rounded to one decimal.
How rankings work
- Each agent is ranked in its main category by overall score. It can be listed in up to two more categories, where it links to its main ranking.
- The overall leaderboard orders every scored agent by overall score.
- Agents without scores are listed but never ranked.
What the numbers mean
Best in class. Sets the standard others are measured against.
Strong. A safe recommendation for most buyers in the category.
Average. Works, with trade-offs you should check against your needs.
Weak. Hard to recommend over the alternatives.
Staying independent
- Rankings are determined by the rubric, not by who's paying.
- Any affiliate links are disclosed and never affect a score.
- Data is researched from primary sources and dated; reviews note when they were last verified.
- Some research is AI-assisted, but scores and verdicts are human-reviewed.
Methodology version 1.0 · last updated June 2026