Subscription Pricing: Examples & Companies

What is it

Subscription Pricing is a pricing model that charges a flat recurring fee — monthly or annual — with no usage component meaningful to the bill.

It is the most predictable model for both sides: the buyer knows the cost in advance, the vendor knows the revenue. That predictability is why it dominated the SaaS era. AI strains it, though, because inference cost varies wildly across users and a flat “all you can use” fee bleeds margin on power users. In the corpus, 161 companies carry a subscription tier — but most pair it with a free tier, a usage meter, or a hard cap, so pure flat pricing is now the minority among AI-native products.

The purest flat-subscription survivors are products with low per-user cost variance or strong structural caps. Midjourney never exposes a per-image meter — it sells tiered GPU-hour bundles at Basic $10, Standard $30, Pro $60, and Mega $120/mo. Wispr Flow holds a flat Pro seat at $12/user/mo (annual) because dictation volume barely varies across users. And per-seat enterprise tools like Jasper and Harvey sell subscription at the seat level rather than the account level.

For the mechanics of picking between flat and metered structures, see our guide on usage-based pricing models; for how subscription drifts toward metering, see subscription-to-usage drift.

How it works

A pure subscription makes one promise — pay $X, get access — and protects margin in one of three ways:

Worked example — Midjourney’s tier ladder. Midjourney sells fast-GPU hours, not images: Basic includes ~3.3 fast hours (~200 images), and each step up buys more (Standard 15 hours, Pro 30, Mega 60). A standard 4-image grid costs ~1 minute of GPU time; video uses ~8x. A heavy user who blows past Basic’s 3.3 hours doesn’t pay per-image overage — they buy up to Standard or run in the free “relax” queue. The bill is flat within a tier; the tier ladder absorbs variance.

Worked example — annual vs monthly. Most subscription AI plans headline the annual rate. Otter Pro is $8.33/user/mo billed annually but $16.99 billed monthly (~51% cheaper on annual). Jasper Pro is $59/seat/mo yearly vs $69 monthly (~20% off). The flat fee stays flat, but the effective rate hinges on billing cadence — a detail buyers should read carefully.

The structural test: if you can use the product 100x more next month and your bill is unchanged, it’s a subscription. The moment a meter or credit pool moves the bill, the corpus reclassifies it as hybrid.

Companies using this

161 companies in the current corpus carry a subscription tier. The flattest sit in consumer creative and chat (Midjourney, Character.ai, Suno, Runway), voice-to-text (Wispr Flow), per-seat productivity (Superhuman, Granola, Otter), enterprise content and legal (Jasper, Harvey), and coding (Cerebras Code). Many others — Notion AI, Gamma, DeepL, ElevenLabs, Perplexity AI — wrap a subscription tier around a free tier or a usage/credit meter. The table below lists each plan’s shape.

Patterns observed

Pure subscription survives where usage variance is low. Wispr Flow has held its flat Pro price since its 2023 launch — one of the most stable headline prices in the AI-tools category while Windsurf and Lovable repriced repeatedly. The same low-variance logic protects Character.ai’s flat c.ai+ and Superhuman’s per-seat mail plans.

Tiered bundles are how subscription absorbs variance without metering. Cerebras applies the tier-ladder idea to coding: Cerebras Code Pro at $50/mo caps at 24M tokens/day and Max at $200/mo at 120M tokens/day — flat prices that hide a daily token ceiling rather than metering overage.

Enterprise subscription means per-seat, often sales-gated. Jasper publishes Pro at $59/seat/mo yearly but custom-quotes its Business tier; Harvey publishes nothing at all, with third-party estimates clustering around ~$1,200 per lawyer per month. Subscription here operates at the seat level, and the account-level price is a negotiation, not a list.

Subscription is increasingly the floor of a hybrid, not the whole model. The dominant move in the corpus is a fixed subscription plus a meter or credit pool. Notion AI bundles AI into per-seat plans but meters advanced agents via Notion credits; Gamma layers a monthly AI-credit allowance (with $6 top-ups for 1,500 credits) onto its per-seat tiers; DeepL pairs per-seat Pro apps with a usage-based translation API. Flat subscription is becoming the “seat” half of a hybrid rather than a standalone model — the subscription-to-usage drift in action.

Counterexamples & variants

The hard counterexample: the pure-usage API with no subscription at all. Developer-facing inference — the token APIs of Anthropic and OpenAI, and Cohere — charge per token with no recurring access fee. These exist precisely because developers reject paying a flat fee for variable machine workloads; a batch job that runs once a quarter shouldn’t carry a monthly seat charge. See pure usage pricing for that model.

The cautionary variant: the flat plan that should have been hybrid. When a single subscription bundles unlimited frontier-model usage, power users turn it unprofitable. Superhuman is the instructive case — it held a flat $30/user/mo for roughly a decade, then in August 2024 repackaged into a $25/$33 two-tier structure and gated its newest AI (Auto Drafts, Ask AI) behind the higher Business tier rather than metering it. That is the subscription-preserving response: split into tiers and gate features, instead of exposing a meter.

The variant that looks flat but isn’t: capped subscription. Otter headlines flat per-seat prices ($8.33 Pro, $19.99 Business, annual) but enforces monthly transcription-minute caps and per-meeting length limits underneath. The bill is flat, but the value is rationed — a subscription-native alternative to metering that keeps the headline simple while protecting inference cost. Buyers should treat any “unlimited” claim on a frontier-model product with suspicion and read the fair-use fine print.

What this means for buyers vs vendors

For buyers

Subscription is the easiest model to budget — but the flat headline hides three things worth checking. First, caps: relax queues, message limits, transcription-minute ceilings, and “fair use” clauses can throttle heavy use (see Midjourney and Otter). Second, billing cadence: the advertised price is usually the annual rate, so confirm the monthly figure before you compare. Third, lurking meters: confirm no metered overage or credit top-up sits behind the flat number, as with Gamma and Notion AI. If your usage is light and steady, a flat plan usually beats a metered one; if it spikes, model the cap before committing.

For vendors

Choose pure subscription only when per-user cost variance is under ~3x, or when you can cap usage without hurting the experience — the Wispr Flow (low-variance dictation) and Midjourney (tier-laddered GPU hours) patterns. Above that threshold, a flat “unlimited” plan on a frontier model is a margin trap. Before reaching for a meter, consider the subscription-preserving moves first: split into tiers and gate premium features (Superhuman), or attach a monthly credit allowance with paid top-ups (Gamma). When those stop working, add a meter and move to hybrid. Our usage invoicing & billing cycles guide covers the mechanics of bolting a meter onto a subscription, and the pricing calculator lets you model the margin impact before you ship it.

Explore this theme in the knowledge graph →