The feature axes that actually differ between assistants
Assistants converge on quality and diverge on everything around it. These are the axes worth checking before committing a team to one.
| Axis | What to check |
|---|---|
| Context length | The real usable limit on your plan, not the headline number |
| File handling | Which formats, how many at once, and whether they persist between chats |
| Web access | Whether it browses live, and whether it cites what it read |
| Code execution | Whether code runs in a sandbox and what that sandbox can reach |
| Memory | Whether it remembers across sessions, and how to inspect and clear that |
| Integrations | Connectors to the tools your team actually uses |
| Data retention | Whether your inputs train the model, and the opt-out path |
| Admin controls | SSO, audit logs, per-seat policy — the reason procurement says no |
Write your twenty real questions first, then run all of them through each candidate. Grading afterwards on impressions reliably picks whichever one you tried most recently.
It depends on the product and the plan, and it changes. Check the current terms for the exact tier you are buying and confirm the opt-out path before rolling it out to a team.