Poe

Poe — Quora’s Multi-Model AI Switchboard (200+ Models, Compute-Points Meter)

Poe (poe.com, Quora Inc., SF/NYC, launched Feb 2023 by Adam D’Angelo) active 2026-08, no discontinuation. It is not a model maker​ — it’s the “browser for AI”: one login, one chat history, one shared compute-points balance across 200+ text/image/video/audio models​ (GPT-5.6-Sol, Claude Opus 5 / Fable 5 / Sonnet 5, Gemini 3 Pro, Grok, Llama, DeepSeek, Kimi-K3, FLUX, Nano Banana 2, Runway, Kling, Veo 3.1, ElevenLabs, Ideogram) plus 1M+ community bots. Category ai-productivity-data; pipeline_stage none. Killer differentiator = side-by-side multi-model comparison + per-bot system prompt + creator monetization + OpenAI-compatible API + Group Chat (200 ppl / 200 models)​ in one tab; not ChatGPT/Claude (single-vendor depth, native Codex/Projects, memory), not OpenRouter (dev-API-first, no consumer chat/bots/group), not Perplexity (cited search, not model router), not Character.AI (roleplay-only). 2026 traps: points don’t roll over​ (free ~300/day down 90% from ~3,000 since Mar 2026; $19.99 Premium cut 1M→660k/mo Jul 2026 at same price); frontier models burn hundreds–thousands pts/msg, video 5k–20k/clip; Knowledge Base feature retired May 18 2026​ (bot RAG killed, files deleted, one-month notice — not a shutdown, write feature-retire); always one hop behind native model releases; Quora-owned → procurement/privacy side-eye; no cross-session memory like ChatGPT.

1. One-line positioning

Poe = the model-comparison cockpit, not a chatbot: sign in at poe.com / iOS / Android / Win / mac → pick Assistant (default router) or any bot → ask GPT-5.6-Sol, flip to Claude Opus 5, run Gemini 3 Pro on the same thread (“compare”) → spawn a multi-bot chat (side-by-side grids) → build a custom bot (system prompt + model pin + optional Knowledge Base pre-May 2026 + monetization price) → publish to marketplace, earn per-message USD from others’ point spend → Group Chat (Nov 2025, 200 participants, each spends own points, mention @GPT5 @Claude in-thread) → Poe Apps (Feb 2025, no-code canvas/app builder) → Script Bots (late 2025, Python workflows chaining models) → Poe API (Jul 2025, OpenAI-compatible, tool-calling for OpenAI/Anthropic/Google, 500 req/min, plugs into Cursor/Cline/n8n). Fits ai-productivity-data; pipeline_stage none (pick model→prompt→compare→bot→app→api). For people who bounce between Claude-for-code and GPT-for-prose and Gemini-for-summary, one $19.99 beats three $20 subs; for single-model loyalists, native app wins on limits/memory/features.

2. Core features

3. Pricing (2026-08, USD — compute-points, no rollover, ~17% off annual)

pricing_type = freemium-plus-six-tier-subscription-plus-shared-compute-points-no-rollover-plus-api; starting_price = $0; free_tier = "yes-but-throttled; ~300 compute pts/day (~15-30 short msgs), light models only (Claude Haiku/GPT-4o-mini class), no frontier GPT/Claude-Opus/Gemini-Pro, no video, cut ~90% from ~3000/day since Mar 2026 unannounced"; pricing_note = "poe-quora-feb-2023-dangelo-active-2026-no-shutdown; points-no-rollover-free-300-day-premium-19.99-660k-mo-cut-from-1m-jul-2026; starter-4.99-10k-day; pro-49.99-1.65m; premium-99.99-3.3m; max-249.99-8.25m; addon-30-per-1m-1yr-expiry; knowledge-base-feature-retired-2026-05-18-files-deleted-not-shutdown-feature-retire; group-chat-nov-2025-200ppl; poe-api-jul-2025-openai-compatible; creator-per-msg-usd-payout-up-to-20-per-subscriber"

4. Access Type

access_type: web-app-plus-ios-plus-android-plus-win-mac-desktop-plus-api

access_display: 🌐 poe.com · 📱 iOS/Android (native, rated pleasant) · 🖥️ Win/mac desktop (native) · 🔌 Poe API (api.poe.com, OpenAI-compatible, tool calling) · no Linux native, no offline, no plugin ecosystem beyond bots/apps

5. Reviews (2026)

Praise: one balance across 200+ models​ (kill alt-tab between ChatGPT/Claude/Gemini), side-by-side compare is genuinely decision-changing for research/coding, custom bot + marketplace + per-message USD payout = distribution without building UI, Group Chat 200/200​ no rival clone, Poe API OpenAI-compatible lowers multi-model dev cost vs OpenRouter+vendor keys, cache discounts (50/90%) reward iterative coding, Starter $4.99 is cheapest frontier toe-dip, opt-in Memory + private chats + provider-no-training clause (Apr 2026) better than raw vendor free tiers.

Gripes: points micromanagement​ (every msg is a transaction, frontier burns hundreds–thousands, video 5k–20k/clip, flow-state breaks), free 300/day + Premium 1M→660k cut​ = quiet ratchet, Knowledge Base retired May 18 2026​ stranded RAG bots (files deleted, 1-mo notice, no export — not shutdown but trust hit), always lags native model launches (no day-1 Codex/Projects/memory parity), Quora parent = procurement side-eye + privacy questions despite no-train clause, custom bots shallower than GPTs/Claude Projects (no fine-tune, no durable file RAG post-May 2026), point UI sometimes mismatches actual remainder, group chat needs every participant on paid plan for frontier.

6. Best for / Not for

7. Competitors