Quick answer

Use Poe to test one task on several models, pick the one that suits each job, save your best instructions in a custom bot, and use group chats for shared work. Watch compute-point costs and verify important claims regardless of how many models agree.

What Poe offers

• Many models in one place: Poe has advertised access to more than 200 models across text, image, video and audio, including models from major AI labs; the exact list varies over time.
• Custom bots: prompt bots (your instructions on top of a model) and server bots (connected to external systems), with creator-made bots shareable in the community.
• Group chats: announced in November 2025, allowing up to 200 other people and multiple models in one conversation.
• Apps and API: Poe has described a way to build small AI apps and an OpenAI-compatible API for developers.
• Compute points: usage is metered in points, with different costs per model and per message.
• Cross-device sync: chats sync across web, desktop and mobile apps.

Using Poe as a testing bench

A simple comparison method

1. Write one clear prompt with the goal, context, constraints and format.
2. Send it to two or three different models, changing nothing.
3. Compare on the criteria that matter: accuracy, completeness, tone, following instructions, and cost in points.
4. Note which model wins for which task type.
5. Reuse that model for similar tasks and re-test occasionally, since models change.

What to compare

Prompting across models

Summarize the attached policy in five bullets for new employees. Then list three ambiguities a new employee might misunderstand. Use plain language, no jargon, and quote the exact policy sentence for each ambiguity.

Good habits:

• Keep prompts identical when comparing so differences reflect the model.
• Put format requirements in the prompt because models differ in default style.
• Ask a second model to critique the first's answer and treat the critique as a lead to check, not a verdict.
• Use cheaper models for drafts and more capable ones for final passes to conserve points.

Workflows

Draft, critique, revise

1. Draft with a fast, inexpensive model.
2. Send the draft to a stronger model with the instruction: "Critique this for weak arguments and unsupported claims. Don't rewrite yet."
3. Revise yourself or ask for edits, then verify facts.

Build a reusable bot

1. Write the instructions you keep retyping, such as "Act as a careful copy editor; flag claims that need sources; never change meaning."
2. Create a prompt bot on the model that performed best in your tests.
3. Test it on three real examples and refine the instructions.

Team brainstorm

1. Start a group chat with teammates.
2. Bring in one text model for ideas and one image model for mood boards.
3. Agree on what to keep, then move the outcome into your normal tools.

Verification

• Agreement between models is not proof; they can share the same mistakes.
• Check facts, quotes and numbers against primary sources.
• Be careful with creator-made bots: instructions and behavior come from the bot's author, so review what a bot does before trusting it with sensitive material.
• Re-run important comparisons when models update.

Common mistakes

• Changing the prompt between models and drawing conclusions about the models.
• Spending points on the most expensive model for trivial tasks.
• Sharing confidential data with community bots.
• Assuming the model name in the picker guarantees identical behavior to the provider's own app.

Limitations and alternatives

• Features and settings may differ from the provider's own apps, and new capabilities can appear on the provider's platform first.
• Point allowances and plan terms change, and heavy use of premium models can use them up quickly.
• If you use one model daily and need its full feature set, its native app may be the better fit.

FAQ

What is Poe?

An app from Quora that provides access to many AI models and bots in one interface.

What are compute points?

Poe's usage meter; models consume different amounts, and allowances depend on plan.

Can I create my own bot?

Yes: prompt bots and more advanced server bots.

Does using multiple models make answers more accurate?

It can reveal disagreements worth checking, but agreement is not proof.