Poe is Quora's AI chat app that puts Anthropic's Claude, OpenAI's GPT, Google's Gemini, xAI's Grok and more than a hundred other models — plus image, video and audio generators — behind one login and one bill. Instead of running separate ChatGPT and Claude subscriptions in separate tabs, you pick a model for the task in front of you and pay for it out of a shared pool of "compute points." It's built for people who bounce between models, compare their answers, or just don't want to pay $20 three times over for tools they each use a few hours a week.
Key Features
- One subscription covers 100-plus models — Claude Opus and Sonnet, GPT-5-series models, Gemini, Grok, DeepSeek, Llama, Mistral, and image/video/audio tools like FLUX, Imagen, Veo and Sora — with no separate account or card for any of them.
- Multi-bot chat lets you @-mention several models in the same thread and read their replies side by side, so you can check a Claude answer against GPT without leaving the conversation.
- Custom bots with knowledge bases: attach PDFs or files, write a system prompt, save it, and reuse it. Prompt bots need no code; server bots hook into Poe's API for anything heavier.
- Native apps on web, iOS, Android, macOS and Windows, with synced history everywhere and desktop keyboard shortcuts for switching models mid-thought.
Screenshots
Why You'll Love Poe
One subscription instead of a stack of them
Run the numbers on paying providers directly: ChatGPT Plus is $20 a month, Claude Pro is $20, Google's AI Pro plan is about $20. That's roughly $60 for three tools, each walled off in its own app. Poe's $19.99 Premium plan covers all three plus Grok, DeepSeek and the image and video models, and you switch between them with a dropdown. If you genuinely move between models during a normal week — Claude for writing, GPT for code, Gemini for long documents — that's the entire pitch, and it holds up.
Multi-bot chat is the part you'll miss elsewhere
Type a prompt, @-mention two or three bots, and Poe fans the question out to all of them and stacks the answers in one thread. It's genuinely useful for fact-checking a shaky response, picking the cleanest phrasing of a paragraph, or watching how a reasoning model and a fast model diverge on the same problem. Group chats add other people to a thread alongside the bots, which is handy for shared research. No native app does this as well.
Plans and pricing
Everything runs on compute points, and different models cost different amounts per message. Here's how the tiers break down at the time of writing (Poe adjusts these allowances fairly often):
- Free — a few hundred points a day, reset every 24 hours, no rollover. Enough to sample the lighter models, not enough for sustained work.
- Starter — $4.99/month or $49.99/year: 10,000 points per day.
- Premium — $19.99/month or $199.99/year: 660,000 points per month (Poe trimmed this from 1,000,000 during 2026).
- Premium Plus — $49.99/month: 1.65 million points per month.
- Pro — $99.99/month: 3.3 million points per month.
- Pro Max — $249.99/month: 8.25 million points per month.
What the points actually buy
Annual billing saves about 17% across the board. Run out mid-period and you can either upgrade a tier or buy add-on points, which run roughly $30 per million and stay valid for about a year. Points do not roll over between periods on any plan — unused daily or monthly points just disappear. Third-party tallies put a long Claude Opus answer in the thousands of points, a Sonnet reply in the low hundreds, and a small model like GPT-4o-mini under ten, though those figures move with message length and Poe's own repricing.
Where the point system bites
The expensive models drain the meter fast. On Premium's 660,000 monthly points, a habit of long back-and-forths with Claude Opus or a top reasoning model can leave you rationing before month's end, and the free tier is now closer to a demo than a usable plan. You also lose the native extras: no ChatGPT Advanced Voice Mode, no Claude Projects or Artifacts exactly as Anthropic ships them, no Gemini Workspace integration, and capped access to Sora and Veo. Poe sometimes trails the native apps by days when a new model launches. And the honest catch — if you only ever touch one model, paying that provider direct gets you the full native experience for the same $20.
How to get started
Install Poe from the App Store or Google Play, or just open poe.com or the macOS/Windows app. Sign in with Google, Apple or email. Pick a bot from the left rail, or browse Explore to find community bots and the newest models. Type a prompt; the point cost for that message shows near the send button. Switch models mid-thread by starting a new message with a different bot or @-mentioning one. To build your own, hit Create, paste a system prompt, attach any reference files, and save. Subscribe from Settings once the free points stop stretching.
Category: AI Assistants
Platform: Web, iOS, Android, macOS, Windows
Price: Free tier; paid plans $4.99–$249.99/month in the US and €5.74–€287.49/month in EU.
Frequently Asked Questions
Yes, there's a free tier. You get a small daily allowance of compute points — a few hundred, reset every 24 hours with no rollover — which is fine for trying the lighter models but runs dry quickly on anything like Claude Opus or a reasoning model. Paid plans start at $4.99/month for 10,000 points a day, with $19.99, $49.99, $99.99 and $249.99 tiers above that. Annual billing takes off about 17%.
No. Every model on Poe runs in the cloud, so the web, desktop and mobile apps all need an internet connection. There's no offline mode and no on-device model. Your chat history syncs back across devices once you're online again.
On mobile it asks for the usual optional ones: photo library and camera access if you want to upload or take images for a prompt, microphone for voice input, and notifications. Apple's App Store privacy label lists data collected and linked to your identity, including purchases, contacts, photos, search history, user ID, usage data and diagnostics. The developer is Quora, Inc. You can use the web version without granting any app-level device permissions.