How 8 real AI products price in 2026

AI Pricing Lab · Updated 2026-07-17 · Prices verified 2026-07-17 and change often; every claim links its source

Cursor: the subscription that is really a prepaid pool

Pro at $20, Pro+ at $60, Ultra at $200, where each plan includes roughly its own price in frontier-model usage, billed at token rates (about $1.25 per million input and $6 per million output, cache reads cheaper), with overage in arrears. Auto mode is unmetered on paid plans; picking a frontier model manually burns the pool (CloudZero's breakdown, June 2026). The design followed the June 2025 repricing that triggered a user revolt and a public CEO apology with refunds. The gotcha for buyers, and the lesson for founders, is model-choice sensitivity: the same task costs zero or dollars depending on a dropdown.

GitHub Copilot: the biggest flat-to-usage migration yet

On June 1, 2026, GitHub moved every Copilot plan to usage-based billing: plans include a monthly allowance of GitHub AI Credits consumed by token usage at published API rates ($10 Pro includes $10, $39 Pro+ includes $39; Business $19 and Enterprise $39 per user). Completions stay unmetered; chat, agents, and CLI burn credits. Developer reaction was loud: "you will get less, but pay the same price." Context makes the move legible: in the flat era Copilot reportedly lost $20 per user per month on average (ARK disputed the numbers). This is the reference case for repricing an installed base, covered in our repricing playbook.

ChatGPT: proof that seats are not dead

Business runs $25 to 30 per user per month with a 2-seat minimum, Enterprise is custom, and heavy usage is handled with credit top-ups on top of seats. Seats remain the easiest thing in software to buy, and OpenAI keeps them as the anchor while venting the cost tail through credits. For founders whose buyers budget in headcount, this is the template: keep the seat, add the valve.

Intercom Fin: outcome pricing, with a contested unit

Fin charges $0.99 per billable outcome (resolution, procedure handoff, disqualification; $9.99 for a qualified lead), minimum 50 outcomes a month. The known gotcha: a "resolution" includes conversations where the customer just goes quiet, so bills can spike without satisfaction rising. Salesforce agreed in June 2026 to acquire the Fin business; pricing was unchanged as of mid-July 2026. Outcome pricing is the strongest value story in AI and the most dispute-prone unit. Benchmarks elsewhere: Zendesk at $1.50 committed / $2.00 pay-as-you-go per automated resolution, HubSpot Breeze at $0.50 per resolved conversation since April 2026.

Replit: effort-based credits, priced after the fact

From July 1, 2026, Replit Agent prices each checkpoint by the time and computation it consumed: simple edits around 5 to 20 credits, complex features 150 to 300 or more, with the price visible only after the checkpoint completes, and post-quota usage charged to the card. Replit is the canonical margin story, the 36 percent to negative 14 percent swing, and effort pricing is its structural fix. The trust cost of after-the-fact pricing is real, and it is the single design choice we would not copy without spend caps and previews.

v0: tokens dressed as credits, with expiry

Vercel's v0 runs Free ($5 of monthly credits), Premium $20 (includes $30 of credits), and team tiers, metering input and output tokens converted to credits across three model tiers; purchased credits expire after a year. It replaced fixed message counts with token metering in 2025, the same direction as everyone else: away from countable units toward cost-tracking ones.

Salesforce Agentforce: two meters, pick one

Agentforce offers roughly $2 per conversation (a 24-hour session) or Flex Credits at $500 per 100k, where a standard action burns 20 credits, about $0.10. The two-meter design exists because enterprises disagree on what unit maps to value, which is the honest admission underneath all AI pricing: the unit is the strategy.

Claude: subscription with usage windows

Anthropic's consumer plans (Pro $20, Max at $100 and $200) meter through rolling 5-hour windows plus weekly caps, with the Max tiers sold as multipliers of Pro's allowance rather than fixed quotas. It is a cap-first hybrid: the subscription anchors the price, the windows protect the margin without a bill that can surprise anyone.

What the eight have in common

• Every model meters somewhere. Unlimited is gone from serious AI products. The choice is only where the meter sits: tokens, credits, outcomes, conversations, or time windows.
• Subscription anchors, usage protects. Six of eight pair a predictable base with a metered component, the hybrid structure from our seats versus usage versus credits comparison.
• Repricing is survivable, opacity is not. The backlash cases (Cursor 2025, Copilot 2026, Replit's after-the-fact pricing) were all about surprise, not the new rates. Publish the meter, alert before the cap, and grandfather with a date.