Claude has the worst pricing – but people want it. Is it worth it?

Value depends on the outcome. While Claude API (Haiku 4.5) costs $0.1 per million tokens for prompt caching (read), its worth is determined by comparing this usage cost to the time saved on complex tasks. If the output reduces human labor hours, even 'expensive' models can show a high net return.

What is the Great AI Repricing of 2022-2025?

The market is shifting from free AI-assisted seats (Ramp, Brex) toward more structured costs. This includes seat-based tiers like Box at $5 per month and granular usage models like Ministral 3B 3.0 at $0.1 per million input or output tokens. Buyers are now prioritizing models that charge for value over those that charge for access.

How do you integrate market data with AI for pattern detection?

Advanced evaluations, such as those used for volatility metrics or flow analysis, typically utilize usage-based models to manage large datasets. Costs for these operations are often calculated per million tokens, such as Mistral's pay-as-you-go input at $0.11 per million tokens, making it essential to track if the analysis results in faster decision-making.

AI Costs & Pricing

Per-Outcome AI Pricing Explained

By Paige Gilmore, Founder, NetLift· Published July 28, 2026· Updated August 18, 2026

Prices checked August 18, 2026 against official vendor pages

Per-outcome AI pricing is a billing model where customers pay for specific results, such as a resolved support ticket or a completed report, rather than paying for user seats or raw data processing. This model shifts the financial risk to the vendor by ensuring costs are only incurred when measurable business value is delivered.

14-day free trial. No credit card required.

Per-outcome pricing is emerging as the preferred model for Finance and Procurement teams who want to move away from speculative software spending. By charging for a finished task instead of a seat license, vendors align their revenue directly with the buyer's realized utility.

This shift addresses the limitations of legacy models. For instance, seat-based pricing ranges from free options like Ramp to $5 per month for Box Business Starter, while usage-based models like IBM Granite cost $0.06 per million input tokens. Per-outcome models bridge this gap by focusing on the 'work' performed rather than the tools used.

Verified seat-based pricing examples

Seat-based example	Price	Unit
Brex — Essentials	$0	per seat month
Amazon Q Business — Lite	$3	per seat month
Otter.ai — Pro (India Special Offer)	$4.17	per seat month
Box — Business Starter	$5	per seat month
Olakai Assistive — Olakai Assistive	$5	per seat month

14-day free trial. No credit card required.

Verified usage-based pricing examples

Usage-based example	Price	Unit
granite-4h-small — Input tokens	$0.06	per 1M input tokens
mistral-small-3-1-24b-instruct-2503 — Input tokens	$0.11	per 1M input tokens
Google Gemma 4 31B — On-Demand Standard Input (US)	$0.14	per 1M input tokens
gpt-oss-120b — Input tokens	$0.16	per 1M input tokens
gpt-5.6-luna — Standard Short Context - Input	$0.2	per 1M input tokens

Why is the market moving away from seat-based pricing?

Traditional seat-based models, such as Amazon Q Business Lite at $3 per seat or Olakai Assistive at $5 per seat, often lead to 'shelfware' where licenses are paid for but rarely used. Procurement teams are increasingly looking for models that link spend to activity. Free tiers from providers like Brex or Ramp have set a baseline expectation that basic AI assistance should be a commodity, forcing specialized AI tools to prove value through outcomes.

14-day free trial. No credit card required.

How does usage-based pricing differ from outcomes?

Usage-based pricing, such as Mistral AI at $0.11 per million tokens or Claude API (Haiku 4.5) at $0.1 per million tokens, charges for the underlying compute. While granular, this model is difficult to budget because the volume of tokens does not always correlate with the quality of the output. Outcome-based pricing simplifies this by charging for a successful result, regardless of how many tokens were required to get there.

NetLift determines if these pricing models deliver a true return by measuring the labor value of time saved against the total cost of adoption. Using a default loaded staff cost of $75 per hour, NetLift calculates the net value of work performed. This methodology looks at objective evidence grades rather than self-estimates, ensuring that an 'outcome' actually translates to reduced labor hours or increased volume. This allows CIOs to decide whether to expand or stop AI spend based on verified financial performance without resorting to employee surveillance.

Frequently Asked Questions

Ready to see your AI return?

14-day free trial. No credit card required.

About the author

Paige Gilmore · Founder, NetLift

Paige Gilmore is the founder of NetLift, the AI Value Management platform that helps organisations measure the cost, savings and return of AI adoption.

Paige Gilmore on LinkedIn

Keep reading

ai-costs

AI Usage Limits and Overage Costs

AI usage is typically billed via monthly seat licenses ranging from $0 to $5 per user, or metered token-based pricing starting at $0.06 per million tokens. Managing costs requires balancing these limits against the verified labour value of time saved to ensure a positive net return.

Read ai-costs

Contact-Sales AI Pricing Explained

Contact-sales AI pricing typically combines low-cost per-seat licenses, ranging from $0 to $5, with variable usage fees starting as low as $0.06 per million tokens. Buyers should prioritize net value by measuring these costs against the loaded labor value of time saved.

Read ai-costs

Monthly vs Annual AI Software Pricing

Monthly AI pricing provides the flexibility to test tools with minimal commitment, while annual pricing typically stabilizes costs for proven workflows. The choice depends on whether the tool uses seat-based fees, like Amazon Q Business, or usage-based token models, like IBM Granite.

Read ai-costs

AI Credits vs Token Billing

Token billing charges for precise data processing volume, typically per million units, while credits or seat-based models provide a fixed monthly cost per user. The choice depends on whether your priority is granular usage control or fixed budget predictability.

Read ai-costs

AI Seat Pricing vs Usage-Based Pricing

Seat pricing offers a predictable fixed monthly cost per user, while usage-based pricing charges for actual consumption, typically measured in millions of tokens. The choice depends on whether your workload is consistent across a set number of users or varies based on automated volume.

Read