For users
Find tools that fit your work
Start with your tasks and requirements. Compare testing conditions and delivery examples in public reports to shortlist Agents and plugins, then check their results in your own setting.
Browse the libraryABOUT
About EvalDock
EvalDock is a cross-ecosystem platform for evaluating and selecting AI Agents. Public reports, evaluation methods and tools help users choose suitable AI assistants and help developers improve Agents and plugins.
From selection to improvement
Use task performance to guide your next decision.
For developers
Find a path to better performance
Use public reports, evaluation methods and tools to inspect delivery performance on your target tasks. Analyze issues using the inputs, outputs and execution records you can obtain, identify improvements and retest under comparable conditions.
Explore evaluation toolsHow we evaluate
Verify delivered results against task requirements and use available evidence to explain execution.
Define tasks and criteria
Select tasks around real needs and specify inputs, configurations and delivery requirements.
Execute and verify
Inspect delivered files and final answers, using task records and execution evidence to check results.
Score and report
Score against task criteria and use component scores and examples to describe performance and improvement opportunities.
Published evaluations
Explore delivered results through scores and examples, within each report’s testing conditions and comparison limits.
Four Office Agents Tested on 75 Tasks
Compare 300 final deliveries from Qianwen Office, WPS Lingxi, WorkBuddy and Doubao Work across office files, spreadsheets, presentations and document questions.
- Chinese report
DSH plugin-configuration evaluation
Explore configuration-level scores, sample sizes and six task examples. Different task compositions do not establish a controlled comparison of plugin effects.
Frequently asked questions
Choosing tools, exploring evaluations and interpreting results.
Where should I start when choosing an Agent for a specific task?
Define your input materials, expected deliverables, environment, acceptable human involvement and costs. Find relevant candidates in the library, then compare report conditions, delivery examples and scope to shortlist Agents or plugins to try in your own setting.
How can developers use EvalDock to improve Agents or plugins?
Use task requirements and delivery examples from public reports, together with evaluation methods, to inspect your product. The tools page links to the EvalDock evaluation tool and BenchDock dataset. Distinguish confirmed delivery defects from causes that still need verification, then retest changes under comparable conditions. Consult each project’s documentation for usage and data scope.
Which evaluations can I read?
The four-office-Agent evaluation covers 75 tasks, with scores and delivery examples by task type. The DSH plugin-configuration report includes configuration scores, sample sizes and six task examples. Both are listed in the results directory; each report states its testing conditions and scope.
How do I judge whether a tool fits my tasks?
Compare the task type, operating conditions and delivery requirements with your own needs, then inspect the case’s execution and verification steps. Configuration scores reflect the tasks included for that configuration; your own context and constraints still matter.
How is Top100 related to EvalDock?
Top100 is EvalDock’s plugin and Skills discovery section for finding candidate tools. Evaluation reports use specific task records to help readers understand how tools and Agents perform.
Explore more of EvalDock
Official information
Top100 is EvalDock’s plugin and Skills discovery section. Inclusion or popularity rankings do not indicate that a tool has passed a capability evaluation.
When citing findings, preserve the report’s tested products, versions, operating conditions and scope.