Product
ChatGPT
OpenAI's consumer chat product, the only LLM application globally to scale past 100 million weekly active users — the largest case of L3a self-serving L4.
1. Core Product / Service
ChatGPT is OpenAI packaging its own 1P Token API (GPT-4o / o1 / o3 / GPT-5 series) into subscription + credit form sold to end users. The tech stack is fully vertically integrated: model weights self-trained, inference self-served (partly on its own Stargate cluster, partly rented from Azure); App / Web / iOS / Android / desktop, Memory, Custom GPTs, Canvas, Advanced Voice, Search, Deep Research, Operator (agent), and other features all stacked on the same subscription.
Four pricing tiers:
- Free — GPT-5 mini rate-limited access, with common restrictions; ad testing underway
- Plus $20/month — high GPT-5 / o3 quota, Advanced Voice, Deep Research limited quota
- Pro $200/month — nearly unlimited GPT-5 / o3-pro, Operator, Deep Research high quota
- Team $25-30/seat·month, Enterprise by negotiation — SSO, long context, no-training commitment, SCIM
2. Target Users & Pain Points
- Individual users (most are Free / Plus): daily writing, info organization, translation, education, code snippets, psychological venting. Pain point is the "search → filter → stitch" loop of search engines + Wikipedia + StackOverflow being too slow.
- Professional users (Pro $200): researchers, lawyers, consultants, deep-report authors — pain point is regular Plus can't finish a long Deep Research.
- Enterprise (Team / Enterprise): compliant deployment + connecting to corporate knowledge base + no-training commitment.
- Developers: don't directly use the ChatGPT line; they go through openai-api (a different product form within the same company).
3. Competitive Landscape
| Product | Pricing | Weekly/Monthly Active (est.) | Differentiation |
|---|---|---|---|
| ChatGPT | $0 / $20 / $200 | 800M+ weekly active (2026 Q1 OpenAI-disclosed) | Brand + model capability + Memory + ecosystem (GPT Store / Operator) |
| claude-ai | $0 / $20 / $100 / $200 | ~30-50M weekly active level, not officially disclosed | Coding strength, long context, Artifacts |
| gemini | $0 / $20 / $200 | ~400M+ monthly active (Google-disclosed) | Embedded in Google One / Workspace / Pixel / Search |
| poe | $5-200/month | ~5-10M monthly active | Multi-model aggregator, not locked to single vendor |
| China's Doubao / Kimi / DeepSeek web | Mostly free | Combined hundreds of millions weekly active in China | Chinese + free |
ChatGPT's brand awareness in Western markets is multiple times the sum of all other LLM applications. OpenAI's brand moat as the "first to scale" currently has no visible challenger.
4. Unique Observations
- Massive markup of retail price vs underlying API cost: the GPT-5 quota given by Plus $20/month, back-calculated from OpenAI's API price, normal user actual token consumption value is usually only $1-5, with 80%+ gross margin perfectly reasonable. Some heavy Pro $200/month users (Deep Research / Operator) can run at a loss, but on average it's still a high-margin product. This is the highest markup multiple along the ai-infrastructure value chain (see framework.md L4 Path 3).
- Typical user token consumption: free users ~50K-200K tokens/month; Plus users ~500K-2M tokens/month; Pro heavy users can hit 50M+ tokens/month (mainly from Deep Research single sessions consuming millions of tokens each).
- API vs Consumer revenue split: per The Information and other media disclosures, OpenAI 2026 ARR is approaching $20B scale, of which ChatGPT subscriptions account for ~70%, API ~25%, enterprise/others remaining. I.e., "selling to end users" is far more profitable than "selling to developers," validating framework.md's Path 3 thickest-value judgment.
- Distribution moat = brand + mindshare occupancy: unlike gemini's "Google ecosystem hard binding" and claude-ai's "developer reputation," ChatGPT's moat is "even people who don't know AI know ChatGPT." This zero-CAC customer acquisition is rare in consumer internet history.
- Data / workflow lock-in: Memory (cross-session user preference memory), Custom GPTs, Project (bound files + system instructions), Canvas (structured documents) jointly form switching cost. Strongest lock-in comes from Memory — the longer used, the better it knows you, and replicating to Claude / Gemini requires starting from scratch.
- L3a → L4 vertical integration textbook: OpenAI is the most complete case of fully verticalizing "self-trained model + self-served inference + self-sold consumer." Anthropic / Google take similar paths but neither is as thorough as ChatGPT (Anthropic API share remains large, Google constrained by Search main business).
5. Financials / Funding
ChatGPT not separately disclosed; all data folded into OpenAI parent:
- 2026 Q1 ARR: ~$20B+ (Reuters, The Information, combined reports, not officially confirmed)
- ChatGPT subscription share: ~70% (estimated)
- Paid subscribers: ~30M+ (Plus + Pro + Team combined; Pro reportedly <1M)
- Weekly active: OpenAI disclosed 800M+ weekly active at end of 2025
- First launch: 2022-11-30
- Milestones: hit 1M users in 5 days (2022-12); hit 100M monthly active in 2 months (2023-01) — still the fastest in consumer internet history.
6. People & Relationships
- Parent: OpenAI (CEO Sam Altman); ChatGPT product lead has changed multiple times
- Key models: GPT-3.5 → GPT-4 → GPT-4o → o1 → o3 → GPT-5 series
- Infrastructure: Microsoft Azure (historical mainstay) + OpenAI's own Stargate cluster (online from 2025+) + Oracle / CoreWeave partial capacity
- Competitors: claude-ai (strongest rival), gemini, poe, China's Kimi / Doubao / DeepSeek
- Related apps: Operator (agent), Sora (video), Advanced Voice
- Aggregator downstream: poe / openrouter both sell GPT series as core SKU
Sources
Related
Last compiled: 2026-05-10