The final Act I tutorial: stop counting generations, start counting the fully-loaded cost of one finished asset. All prices verified as of mid-2026; check the vendor page before you budget.
Three pricing models. Flat seats (rewards iteration, punishes idle seats), credit/token metering (rewards discipline, punishes exploration via "credit anxiety"), and the industry drift toward consumption pricing: PYMNTS reports SaaS moving off flat subscriptions, with 63% of enterprises blowing past AI budgets by 30%+ in year one.
Flat-seat tools. ChatGPT Free/$8 Go/$20 Plus/$100–$200 Pro/~$25–30 Business/Enterprise custom (reportedly $40–60/seat at scale). Claude Pro ~$20, Max $100/$200, Team $25/$125. Gemini folded into Workspace, base prices rose 17–22%. Jasper $59–69, Copy.ai $49, Writer $29 to $500K+ enterprise.
Credit-metered. Canva $15 (500 no-rollover credits), Midjourney $10–120 by fast-GPU-hour, Firefly, Runway (625 credits ≈ 25 sec), ElevenLabs (1 credit/char, the rare tool that rolls credits over).
The API layer (Anthropic, OpenAI) explains the draft-vs-final logic: a 25x span from nano to flagship output.
Cost per finished asset (Sight AI): a $100/mo tool making 20 articles at 90 min edit each = ~$85/article, subscription just 6% of true cost. Model routing (70/20/10) cuts spend 60–80%. Hidden costs: wasted seats/tool sprawl, credit expiry, commercial-use floors, and OpenAI's soft spend cap that no longer hard-stops the bill.
News: OpenAI declares itself an ad business at Cannes; HubSpot Agent for M365 Copilot; ActiveCampaign brand memory; Google DSA auto-upgrade to AI Max (Feb 2027); Claude for Slack; and the SparkToro zero-click study (68% no-click).