AI Credits Trend

Companies using credit or token systems to meter AI usage. This cuts across `ai_packaging` values — credit metering is an implementation detail that appears in bundled, add-on, and separate-sku companies alike. No companies in the dataset use a literal `ai_packaging: credit-lane` tag; this page groups by pricing mechanic rather than tag value.

Companies Using Credit/Token Metering

Bundled with credit limits

• cursor — Pro/Pro+/Ultra tiers defined by AI request limits (500/unlimited fast, unlimited slow). Credits gate depth, not access.
• openai — pure per-token usage pricing. No credits per se, but token metering is the entire model.
• anthropic — seat plans with usage limits that increase by tier. API is pure per-token. Enterprise seats now unbundled: $20/user/mo flat, all usage billed separately at standard API rates.
• bolt — token-metered AI app builder. Plans differ by token allowance.
• elevenlabs — character-based credits for AI voice generation, increasing by plan.
• veed — AI video editing with credit limits that increase per tier (`ai_packaging: tiered`).
• microsoft — Copilot Credits at $0.01/credit for agent workloads via Copilot Studio. Dynamics 365 bundles 1,000 credits/seat/mo with pay-as-you-go overage.
• github — shifting to AI Credits (effective June 1, 2026). Plans include credits equal to price ($10/$19/$39). Code completions remain free; chat, agents, premium models consume credits.

Add-on with credits

• notion — Custom Agents add-on priced at $10 per 1,000 Notion credits. Standard AI (editor, chat, search) remains unlimited.
• salesforce — Agentforce uses flex-credits system ($0.10/action, 20 credits per action). 100K free credits for Enterprise Edition via Salesforce Foundations.
• sap — AI Units (~7 EUR/unit) consumed by Joule Premium features. Agent tiers: Basic (0.005/step), Standard (0.01/step), Advanced (0.025/step). No rollover.
• servicenow — Assist Credits for Now Assist consumption, overage at $0.015-0.04/assist.
• atlassian-jira — Rovo credits: 10 per Chat/Agent request, 100 per Deep Research. Virtual Service Agent: $0.30/conversation after 1,000 free/mo.
• figma — AI credit enforcement began March 2026. Free: 500/mo. Enterprise: 4,200/seat/mo. Monthly pool: $120/mo for 5,000 credits (~2c/credit).

Separate SKU with credits

• clickup — Brain ($9/user/mo) and Everything AI ($28/user/mo) plans include credits, plus PAYG credit purchasing.
• dialpad — AI Agent tab with conversation-credit pool. Credits consumed only when AI delivers value.

Usage-based AI infrastructure

• deepgram — per-audio-minute pricing for speech AI
• fireworks — per-token inference pricing
• replicate — per-second GPU pricing for model inference
• huggingface — usage-based for inference endpoints
• databricks — DBU-based consumption for AI workloads
• snowflake — credit-based consumption including AI/ML features

Market Adoption

PricingSaaS 500 Index [PricingSaaS 500 Index — no public URL]: 79 of 500 companies (15.8%) now offer a credit model, up 126% YoY from 35 at end of 2024. Credit adoption is accelerating fastest in enterprise software (Microsoft, SAP, Salesforce, ServiceNow) where AI agent workloads make per-seat economics unsustainable.

Credits are vendor-friendly but customer-hostile — they obscure unit costs and make cross-vendor comparison nearly impossible. Every vendor invented a different currency: Copilot Credits, AI Units, Flex Credits, Assist Credits, Notion credits. The counter-trend: windsurf dropped credits entirely in March 2026 for quota-based daily/weekly limits, suggesting some companies think the complexity isn't worth the lock-in.

The Pattern

Credit systems solve the fundamental tension of AI packaging: customers want unlimited access, companies face variable compute costs. Credits are the compromise.

1. Bundled access + metered depth — cursor bundles AI in every plan but limits fast requests. Pro gets 500 fast, Ultra gets unlimited. Access is free; intensity costs money.
2. Credits as a new value metric — notion's credit system ($10/1,000) creates a separate axis of monetization orthogonal to per-seat pricing. Seats control who; credits control how much.
3. Outcome-based credits — dialpad only consumes credits when AI "delivers value." intercom charges $0.99 per resolution. The unit of consumption shifts from input (tokens, compute) to output (results, resolutions).
4. PAYG as safety valve — clickup offers both plan-based credit pools and pay-as-you-go purchasing. This lets power users burst beyond plan limits without forcing everyone to a higher tier.

The prediction: as AI costs decline, credit pools will get larger per tier and eventually fold into base pricing. The companies currently using credits are building the metering infrastructure they'll need if usage-based AI becomes the dominant model.

See also: ai-bundling, ai-add-ons, no-ai