Poe

Quora's "consumer-side LLM aggregator" — one subscription to access GPT / Claude / Gemini / hundreds of models, with a compute-points economy that recycles costs.

1. Core Product / Service

Poe ("Platform for Open Exploration") was launched by Quora in early 2023, positioned as a consumer-side multi-model chat App. The technical stack pulls 100+ models / agents such as chatgpt / claude-ai / gemini / Mistral / DeepSeek / Flux / Suno / Runway into one unified UI + one subscription.

Core mechanisms:

• Compute Points economy: Poe attaches a compute points price to every message (dynamically priced by underlying model cost), each subscription tier corresponds to a monthly compute-points quota
• Bot ecosystem: users and creators can build "custom bots" on top of existing models + system prompts, publish them publicly; bot creators get a compute-points revenue share
• Multimodal: image, video, and music generation models are also connected (Flux / Imagen / Veo / Suno, etc.), not limited to text

Subscription tiers:

• Free — tiny compute points / day
• $5/mo (Lite, launched 2024) — entry tier, fits non-heavy users
• $20/mo — aligned with ChatGPT Plus / Claude Pro pricing, but usable across all models
• $50/mo, $100/mo, $200/mo — heavy tiers, scaling compute points

2. Target Users & Pain Points

• Users who don't want to be locked into a single vendor: use Claude for code today, Gemini for image analysis tomorrow, Veo / Sora for video the day after — one subscription for all those scenarios.
• Multi-model prompt engineers / creators: need to run GPT / Claude / Gemini side-by-side to compare outputs; Poe is one of the few products supporting multi-bot side-by-side comparison.
• Budget-conscious light multi-tool users: $5 Lite is much cheaper than ChatGPT Plus for people who use LLMs occasionally.
• Image / video / music generation samplers: subscribing separately to MidJourney / Runway / Suno is $10-20+ each; Poe bundles them — the experience may not be best-in-class, but the price is friendly.

3. Competitive Landscape

Poe's differentiation: it is the only consumer product stacking "subscription + creator economy + multi-model" three layers together. OpenRouter also aggregates but targets developers (API); Poe sells to end users who don't write code.

4. Unique Observations

• Aggregator-as-product model: Poe essentially takes openrouter's "wholesale routing" model (framework.md L4 Path 1) and turns it directly into a chatgpt-style consumer App (L4 Path 3) — a cross-layer hybrid. This is a non-typical route: Path 1 + Path 3 crossbreed.
• Compute Points is a piece of clever pricing engineering: underlying model token prices are highly variable (Claude Opus is 100x+ more expensive than Gemini Flash), making it hard to amortize under a fixed subscription. Poe uses compute points to pass volatility through to users: want to use Opus, burn points fast; want to save, use cheaper models. This mechanism also means Poe doesn't immediately fall into the red when a single model raises prices.
• Retail price vs API cost: the $20 / month tier provides roughly ~1M compute points. With Poe's points → token conversion (high-end models ~ several hundred points per message), the $20 user's actual underlying token raw API cost is typically $5-15, leaving 25-75% gross margin. But heavy users running top-tier Opus / o3-pro can easily go upside-down — Poe relies on compute-point caps to force a stop-loss.
• Typical user token consumption: Lite $5 user ~50K-200K tokens/month; $20 user ~500K-2M tokens/month (comparable to ChatGPT Plus but across multiple models); $200 heavy user can hit 50M+.
• Distribution moat = Quora parent site + multi-model selection: Quora still has hundreds of millions of monthly active users globally; Poe is hosted on the quora.com subdomain and pushed via the App, with zero customer acquisition cost. But this is a weak moat, because Quora itself is being drained by ChatGPT in the LLM era.
• Data / workflow lock-in is weak: user chat history, custom bots are in Poe, but prompts are text and bot config is system prompt + model selection — exporting to ChatGPT GPTs / Claude Projects isn't difficult. Poe's lock-in mainly rests on the switching cost of "changing brand means resubscribing", not on data.
• Bot creator economy: Poe pays creators a compute-points share (convertible to USD), but the number of creators actually making money is limited, nowhere near GPT Store / Discord-scale ecosystems.
• Strategic predicament: as OpenAI / Anthropic / Google each strengthen their own Apps, Poe is being squeezed — the value of "multi-model choice" it offers users is depreciating (each vendor's App is getting stronger). Poe's long-term survival depends on "whether the Quora mothership traffic is still there".

5. Financials / Funding

Poe is not separately disclosed:

• Parent Quora: valuation ~$2B (2017 Series D, no recent update)
• Poe revenue: The Information reported in late 2024 an ARR in the $40-50M range, continued growth in 2025 but unofficial
• Paying users: undisclosed, estimated in the millions
• Launch: 2023-02
• Milestones: subscription launched 2023; compute points + creator share added 2024; video / music models added 2024-2025

6. People & Relationships

• Parent: Quora (CEO Adam D'Angelo — also an OpenAI early board member, one of the directors focal during the Sam Altman firing storm in 2023-11)
• Founder / Head: Adam D'Angelo
• Underlying model supply: chatgpt (OpenAI), claude-ai (Anthropic), gemini (Google), Mistral, Meta Llama, DeepSeek, Moonshot Kimi, Stable Diffusion / Flux, Runway, Suno, etc.
• Competitors: openrouter (developer-side), individual ChatGPT / Claude / Gemini Apps, You.com
• Ecosystem: Adam D'Angelo's relationship with OpenAI is key to Poe being able to get top-tier models concurrently

Sources

• https://poe.com/about (2026-05-10)
• https://help.poe.com/hc/en-us/articles/4423666220183 (2026-05-10)
• https://creator.poe.com/docs/economy (2026-05-10)
• https://www.theinformation.com/articles/quora-poe-revenue (2026-05-10)